chore: merge deep-agent into main

[NP-compete]

Summary

• Merges the deep-agent development branch into main
• Brings all accumulated features and fixes

Closes #113

Summary by CodeRabbit

• New Features
  • Added configurable branding, themes, feature flags, announcements, settings, memories, custom rules, and tool approvals.
  • Added searchable chat history, chat management, exports, feedback, task progress, artifacts, debugging, and keyboard shortcuts.
  • Added human-in-the-loop approvals, MCP authentication, session handling, rate-limit messaging, and automatic recovery.
• Accessibility
  • Improved WCAG support, keyboard navigation, focus states, semantic controls, and screen-reader announcements.
• Reliability & Operations
  • Improved streaming recovery, error handling, health indicators, configuration updates, security validation, and observability.
  • Added Redis-backed persistence and Kubernetes/OpenShift deployment support.
  • Added automated testing and dependency, license, vulnerability, and container checks.

[coderabbitai]

📝 Walkthrough

Walkthrough

The PR adds a deep-agent application platform with runtime configuration, Redux state, streaming and HITL workflows, PatternFly UI, server proxying, authentication, OPA, Redis sessions, deployment resources, CI workflows, and automated tests.

Changes

Application platform

Layer / File(s)	Summary
Repository automation and deployment .github/*, .gitlab-ci.yml, Containerfile, Makefile, compose.yml, config/*, deployment/*	Adds repository templates, CI and release workflows, container scanning and signing, runtime configuration, OPA policy, Redis composition, and Kubernetes or OpenShift resources.
Frontend application experience src/frontend/*, src/components/*, e2e/*	Adds Redux state, configuration and branding, PatternFly layouts, settings, chat streaming UI, HITL controls, task panels, error recovery, accessibility behavior, and unit or E2E coverage.
Server integration and validation src/server/*, README.md, env.template, package.json, vite.config.ts, vitest.config.ts	Adds settings-driven Fastify startup, authentication, proxy streaming, tracing, OPA, Redis sessions, runtime routes, caching, security controls, and server validation tests.

Estimated code review effort: 5 (Critical) | ~120 minutes

Possibly related PRs

• redhat-data-and-ai/template-ui#99: Changes the same streaming components for duplicate tool-call responses.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title clearly states that the PR merges the deep-agent branch into main.
Linked Issues check	✅ Passed	The PR objective and changes implement issue #113 by merging deep-agent into main with its accumulated features and fixes.
Out of Scope Changes check	✅ Passed	The broad changeset is consistent with the stated branch merge objective and contains no identifiable unrelated code changes.
Docstring Coverage	✅ Passed	Docstring coverage is 89.47% which is sufficient. The required threshold is 80.00%.

✨ Finishing Touches 💡 1

🛠️ Fix failing CI checks 💡

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch deep-agent

---

_{Comment @coderabbitai help to get the list of available commands.}

[coderabbitai]

Actionable comments posted: 2

Note

Due to the large number of review comments, Critical severity comments were prioritized as inline comments.

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

.github/workflows/ci.yml (1)

1-77: 🔒 Security & Privacy | 🟠 Major | ⚡ Quick win

Add an explicit permissions: block.

No job in this workflow needs more than read access to repository contents. Without a permissions: key, the GITHUB_TOKEN uses the repository or organization default, which can be broader than necessary. Add a workflow-level permissions: contents: read to apply least privilege across all jobs.

🔒️ Proposed fix

 name: CI
 
+permissions:
+  contents: read
+
 on:
   push:
     branches: [main, deep-agent]

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.github/workflows/ci.yml around lines 1 - 77, Add a workflow-level
permissions block near the top-level name/on configuration, granting only read
access to repository contents with contents: read. Apply this setting across all
jobs without changing the existing job steps or other permissions.

Source: Linters/SAST tools

🟡 Minor comments (43)

.github/workflows/ci.yml-14-14 (1)

14-14: 🔒 Security & Privacy | 🟡 Minor | ⚡ Quick win

Set persist-credentials: false on actions/checkout steps that don't need git push access. None of these jobs push commits or tags, so the checked-out credential should not persist in the local git config for the rest of the job. actions/checkout defaults to persist-credentials: true, which leaves the GITHUB_TOKEN accessible to subsequent steps, including third-party code executed by npm ci/npm run.

• .github/workflows/ci.yml#L14-L14: add with: persist-credentials: false to the checkout in the lint job.
• .github/workflows/ci.yml#L26-L26: add with: persist-credentials: false to the checkout in the unit-test job.
• .github/workflows/ci.yml#L38-L38: add with: persist-credentials: false to the checkout in the build job.
• .github/workflows/ci.yml#L60-L60: add with: persist-credentials: false to the checkout in the e2e job.
• .github/workflows/build-base-image.yml#L35-L36: add with: persist-credentials: false to the "Checkout code" step.
• .github/workflows/dependency-review.yml#L18-L19: add with: persist-credentials: false to the "Checkout code" step.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @.github/workflows/ci.yml at line 14, Set persist-credentials to false on
every actions/checkout step in the lint, unit-test, build, and e2e jobs at
.github/workflows/ci.yml lines 14, 26, 38, and 60, plus the “Checkout code”
steps at .github/workflows/build-base-image.yml lines 35-36 and
.github/workflows/dependency-review.yml lines 18-19. No checkout step requires
push access.

Source: Linters/SAST tools

Makefile-34-39 (1)

34-39: 🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Do not terminate unrelated developer processes.

Lines 37 and 38 match every current-user process whose command line matches node.*dist/server or npm.*start. make clean can stop another checkout or service. Track the PID started by make local, then terminate only that PID. Otherwise, remove these pkill commands.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@Makefile` around lines 34 - 39, Update the clean target to avoid terminating
unrelated developer processes: remove the broad pkill commands, or replace them
with termination of only the specific process PID recorded when make local
starts the service. Preserve cleanup of containers, volumes, node_modules, and
dist.

README.md-102-104 (1)

102-104: 📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Add a language identifier to the validation-error block.

Line 102 triggers markdownlint rule MD040. Mark the output as text.

Proposed fix

-```
+```text
 Config validation error: branding.colors.light.primary must be a valid hex color (got 'not-a-color')

</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In @README.md around lines 102 - 104, Update the validation-error fenced code
block in README.md to specify the text language identifier, changing the opening
fence to ```text while preserving the existing error output.

</details>

<!-- cr-comment:v1:251974e52f4714393c1a8f33 -->

_Source: Linters/SAST tools_

</blockquote></details>
<details>
<summary>src/server/utils/settings.ts-220-245 (1)</summary><blockquote>

`220-245`: _🔒 Security & Privacy_ | _🟡 Minor_ | _⚡ Quick win_

**Reject prototype keys in `deepMerge`.**

`result[key] = oVal` assigns any key that appears in the parsed YAML. `js-yaml` keeps `__proto__` as a mapping key, so a config file that contains a `__proto__` mapping pollutes `Object.prototype` through this assignment. The config file is operator-supplied, so this is a hardening gap rather than a remote attack path, but the guard costs one line.

<details>
<summary>🔒 Proposed fix</summary>

```diff
   const result = { ...base } as Record<string, unknown>;
   for (const key of Object.keys(override)) {
+    if (key === "__proto__" || key === "constructor" || key === "prototype") continue;
     const bVal = result[key];
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/server/utils/settings.ts` around lines 220 - 245, Update deepMerge to
reject prototype-related keys, especially __proto__, before reading or assigning
override entries. Keep normal configuration keys and recursive object merging
unchanged, while ensuring rejected keys cannot reach result[key] or mutate
Object.prototype.
```

</details>

<!-- cr-comment:v1:6a987e4649aebc671ff47e63 -->

</blockquote></details>
<details>
<summary>src/server/router/client.router.ts-73-79 (1)</summary><blockquote>

`73-79`: _🔒 Security & Privacy_ | _🟡 Minor_ | _⚡ Quick win_

**Escape `agentName` before you interpolate it into the HTML.**

`getAgentName()` in `src/server/utils/settings.ts` (Line 539-557) reads `name` from the agent `/info` response. The value goes into `<title>` without escaping. An agent that returns `</title><img src=x onerror=...>` injects markup into every page. The CSP `script_src: ["'self'"]` blocks inline scripts, so the impact is markup injection rather than script execution, but the value still needs escaping.

<details>
<summary>🔒 Proposed fix</summary>

```diff
+const escapeHtml = (s: string) =>
+  s.replace(/&/g, "&amp;").replace(/</g, "&lt;").replace(/>/g, "&gt;")
+   .replace(/"/g, "&quot;").replace(/'/g, "&`#39`;");
+
```

Then use it in the template:

```diff
-    <title>${agentName}</title>
+    <title>${escapeHtml(agentName)}</title>
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/server/router/client.router.ts` around lines 73 - 79, Escape the value
returned by getAgentName before interpolating agentName into the HTML template
in the client router response. Use the project’s existing HTML-escaping utility
if available, and preserve the escaped value in the title while leaving the
surrounding page markup unchanged.
```

</details>

<!-- cr-comment:v1:151ef33e899f961b0d887f67 -->

</blockquote></details>
<details>
<summary>src/server/utils/config-edge-cases.test.ts-86-95 (1)</summary><blockquote>

`86-95`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Test the missing implicit config path**

With `UI_CONFIG_PATH` unset, `getSettings()` loads and merges `config/ui/settings.yaml`. This test therefore checks the shipped YAML values, not the missing-file fallback. Rename the test or inject/mock the default path so it uses a nonexistent temporary file. Do not set `UI_CONFIG_PATH`; missing explicit paths throw.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/server/utils/config-edge-cases.test.ts` around lines 86 - 95, Update the
test around getSettings so it genuinely exercises the missing implicit config
path: keep UI_CONFIG_PATH unset and inject or mock the default config path to a
nonexistent temporary file. Rename the test to reflect the scenario, while
preserving the existing default-value assertions; do not set UI_CONFIG_PATH
because missing explicit paths must still throw.
```

</details>

<!-- cr-comment:v1:2f397cfe2b1cc2bb7a826b22 -->

</blockquote></details>
<details>
<summary>src/server/plugins/trace.plugin.ts-23-28 (1)</summary><blockquote>

`23-28`: _🔒 Security & Privacy_ | _🟡 Minor_ | _⚡ Quick win_

**Validate the `x-trace-id` fallback value.**

The `traceparent` path is validated by `TRACEPARENT_RE`, but the `x-trace-id` path accepts any non-empty string. The value is stored on the request, written to every response header, and reflected into logs. Restrict it to a safe charset and a maximum length, and generate a UUID otherwise.

<details>
<summary>🛡️ Proposed fix</summary>

```diff
     if (!traceId) {
       const xt = request.headers["x-trace-id"];
-      traceId =
-        typeof xt === "string" && xt.length > 0 ? xt : randomUUID();
+      traceId =
+        typeof xt === "string" && /^[\w.-]{1,64}$/.test(xt)
+          ? xt
+          : randomUUID();
     }
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/server/plugins/trace.plugin.ts` around lines 23 - 28, Update the
x-trace-id fallback logic in the trace plugin around
request.headers["x-trace-id"] to validate incoming values against a safe
character set and maximum length before accepting them. Reuse the existing
randomUUID fallback for missing or invalid values, while preserving the current
traceparent validation path and subsequent request-header assignment.
```

</details>

<!-- cr-comment:v1:3a798b90d202185f9d823d7b -->

_Source: Linters/SAST tools_

</blockquote></details>
<details>
<summary>src/server/server.ts-163-203 (1)</summary><blockquote>

`163-203`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Update the comparison baseline after each reload.**

`cfg` is captured once, before the watcher starts. `watchConfig` calls `resetSettings()` and `getSettings()`, so `newSettings` is a new object on every reload, but `cfg` still holds the original snapshot. After the second reload, the diff is computed against the original configuration instead of the previous one. The watcher then reports settings as changed that did not change in that reload, and it misses a value that was reverted.

<details>
<summary>🐛 Proposed fix</summary>

```diff
 export function startConfigWatcher(configPath: string, server: FastifyInstance) {
-  const cfg = getSettings();
+  let cfg = getSettings();
   const cleanup = watchConfig(configPath, (newSettings) => {
     server.log.info('[ConfigWatcher] Settings reloaded');
@@
     if (autoApplied.length > 0) {
       server.log.info({ settings: autoApplied }, '[ConfigWatcher] Settings applied without restart:');
     }
+
+    cfg = newSettings;
   });
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/server/server.ts` around lines 163 - 203, Update startConfigWatcher so
its comparison baseline is mutable and replaced with newSettings after each
watchConfig reload. Perform all restartRequired and autoApplied comparisons
against the previous snapshot, then assign the reloaded settings as the baseline
before the callback completes, preserving accurate change and revert detection
across successive reloads.
```

</details>

<!-- cr-comment:v1:cc38e4df5d9ae2e88dcd453c -->

</blockquote></details>
<details>
<summary>src/server/utils/opa.ts-108-111 (1)</summary><blockquote>

`108-111`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Handle non-string policy results.**

`String(d)` produces `"[object Object]"` when a Rego rule yields an object, for example `{"msg": "..."}`. The violation list is returned to clients by `/config/compliance`. Extract a message field when the entry is an object.

<details>
<summary>🐛 Proposed fix</summary>

```diff
-      return denials.map((d) => ({ message: String(d) }));
+      return denials.map((d) => ({
+        message:
+          typeof d === "string"
+            ? d
+            : typeof (d as { msg?: unknown })?.msg === "string"
+              ? (d as { msg: string }).msg
+              : JSON.stringify(d),
+      }));
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/server/utils/opa.ts` around lines 108 - 111, Update the denial mapping in
the OPA result handling to extract the message field from object entries before
converting values to strings, so objects such as {"msg":"..."} return their
meaningful message rather than "[object Object]". Preserve string handling for
primitive entries and the existing response shape used by /config/compliance.
```

</details>

<!-- cr-comment:v1:9c0a3571ce0fa36a35f3bd90 -->

</blockquote></details>
<details>
<summary>src/frontend/components/InterruptBanner.tsx-136-141 (1)</summary><blockquote>

`136-141`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Report a blocked popup to the user.**

`window.open` returns `null` when the browser blocks the popup. The code ignores the return value, so `connecting` clears and no error appears. The user sees an unchanged `Authenticate` button and has no explanation.





<details>
<summary>🐛 Proposed fix</summary>

```diff
-      window.open(body.authorize_url, 'mcp-oauth', 'width=600,height=700');
+      const popup = window.open(body.authorize_url, 'mcp-oauth', 'width=600,height=700');
+      if (!popup) {
+        throw new Error('The browser blocked the authentication window. Allow popups and try again.');
+      }
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/InterruptBanner.tsx` around lines 136 - 141, Update
the popup-opening flow in InterruptBanner to capture the return value of
window.open and treat a null result as a connection failure. Set the same
user-visible connect error used by the catch path before clearing connecting,
while preserving successful popup behavior.
```

</details>

<!-- cr-comment:v1:7b155646cce37eadaa102c54 -->

</blockquote></details>
<details>
<summary>src/frontend/hooks/useDataStream.tsx-209-213 (1)</summary><blockquote>

`209-213`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Guard the last-message update against an empty or non-AI message list.**

`isStreamingTokensRef.current` stays `true` for the rest of the token stream. `setMessages` is returned from the hook at Line 244, so a consumer can replace the message list while the stream runs. If the list becomes empty, `last` is `undefined` and `last.content` throws. If the last entry is a human message, the token text is appended to the wrong message.





<details>
<summary>🐛 Proposed fix</summary>

```diff
                 setMessages(prev => {
                   const last = prev[prev.length - 1];
+                  if (!last || last.type !== 'ai') return prev;
                   const updated = { ...last, content: ((last.content as string) || '') + content };
                   return [...prev.slice(0, -1), updated];
                 });
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/hooks/useDataStream.tsx` around lines 209 - 213, Update the
setMessages callback in useDataStream to handle an empty list and verify the
last message is an AI message before appending streamed content. Return the
existing message list unchanged when no valid AI message is present; otherwise
preserve the current append behavior.
```

</details>

<!-- cr-comment:v1:6af79888017e6800e3552580 -->

</blockquote></details>
<details>
<summary>src/frontend/services/agent-rest.ts-131-143 (1)</summary><blockquote>

`131-143`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**`response.json()` sits outside the `try` block, so a malformed body rejects the promise.**

The doc comment and the `catch` at Line 127 state that request failures and invalid payloads return an empty list. The `try` covers only the `authenticatedFetch` call. If the endpoint returns HTTP 200 with a non-JSON body, `response.json()` throws and the rejection propagates to the caller in `src/frontend/components/layout/AppLayout.tsx` at Line 119. The behavior then differs from `getThreadState`, which wraps the parse.





<details>
<summary>🐛 Proposed fix</summary>

```diff
   if (!response.ok) return [];
 
-  const results = await response.json();
-  if (!Array.isArray(results)) return [];
+  let results: unknown;
+  try {
+    results = await response.json();
+  } catch {
+    return [];
+  }
+  if (!Array.isArray(results)) return [];
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/services/agent-rest.ts` around lines 131 - 143, Move the
response parsing and result validation in the thread-fetching method into the
existing try block that currently handles authenticatedFetch, so response.json()
failures are caught and return an empty list through the existing catch path.
Preserve the current filtering and mapping behavior for valid array payloads,
matching getThreadState’s error-handling behavior.
```

</details>

<!-- cr-comment:v1:7fd3e55ee98b454614821fba -->

</blockquote></details>
<details>
<summary>e2e/accessibility/wcag.spec.ts-158-158 (1)</summary><blockquote>

`158-158`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Avoid the undeclared `axe-core` type import.**

Declare `axe-core` in `devDependencies`, or derive the type from `AxeBuilder['analyze']`. The current import can resolve to an unrelated hoisted version or fail under strict dependency layouts.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@e2e/accessibility/wcag.spec.ts` at line 158, Update formatViolations to avoid
the undeclared axe-core type import by deriving the violation type from
AxeBuilder['analyze'] or by declaring axe-core as a devDependency; prefer the
existing AxeBuilder type path and preserve the function’s current behavior.
```

</details>

<!-- cr-comment:v1:4ec3590816a9ee2617a855a2 -->

</blockquote></details>
<details>
<summary>e2e/chaos/resilience.spec.ts-257-269 (1)</summary><blockquote>

`257-269`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Pass an explicit argument before the timeout options.**

`page.waitForFunction` takes `arg` as its second parameter and `options` as its third. Add `undefined` before the options object in both calls at Lines 257-269 and 388-393.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@e2e/chaos/resilience.spec.ts` around lines 257 - 269, Update both
page.waitForFunction calls in the resilience test to pass an explicit undefined
argument before the timeout options object, preserving the existing predicate
and timeout configuration.
```

</details>

<!-- cr-comment:v1:087f35bc39f86497574c4fc9 -->

</blockquote></details>
<details>
<summary>src/frontend/hooks/useStreamingAPI.ts-596-611 (1)</summary><blockquote>

`596-611`: _🔒 Security & Privacy_ | _🟡 Minor_ | _⚡ Quick win_

**Handle empty `action_requests` interrupts**

When `action_requests` is empty, `ChatPage.tsx` returns before resuming, and `AIMessageRenderer` renders no approval controls. Clear or dismiss the pending interrupt, or resume it explicitly, so the run does not remain pending.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/hooks/useStreamingAPI.ts` around lines 596 - 611, Update
checkAndAutoApprove to explicitly handle an empty interruptValue.action_requests
array by clearing or dismissing the pending interrupt, or resuming it
immediately. Ensure this path does not leave the run pending while preserving
the existing approval decision behavior for non-empty requests.
```

</details>

<!-- cr-comment:v1:739a02635bcb5da527949232 -->

</blockquote></details>
<details>
<summary>src/frontend/components/InterruptBanner.test.tsx-134-141 (1)</summary><blockquote>

`134-141`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Unstub globals after each MCP auth test.** `vi.restoreAllMocks()` does not undo `vi.stubGlobal()`, so `fetch` and `open` remain replaced after this block. Add `vi.unstubAllGlobals()` to `afterEach`; each `beforeEach` creates new `vi.fn()` values, so queued implementations do not carry between these tests.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/InterruptBanner.test.tsx` around lines 134 - 141,
Update the afterEach cleanup in the InterruptBanner test setup to call
vi.unstubAllGlobals() alongside vi.restoreAllMocks(), ensuring the fetch and
open stubs created by beforeEach do not leak between MCP auth tests.
```

</details>

<!-- cr-comment:v1:b766b7b79f532f9153a32f26 -->

</blockquote></details>
<details>
<summary>src/frontend/services/export-chat.ts-35-45 (1)</summary><blockquote>

`35-45`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Defer `URL.revokeObjectURL` until after the download starts.**

Line 44 revokes the blob URL in the same synchronous task as `a.click()`. Some browsers (notably Safari) then cancel the download because the URL is already invalid when they fetch it. Revoke on a later task instead.

<details>
<summary>🛠 Proposed fix</summary>

```diff
   document.body.appendChild(a);
   a.click();
   a.remove();
-  URL.revokeObjectURL(url);
+  setTimeout(() => URL.revokeObjectURL(url), 0);
 }
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/services/export-chat.ts` around lines 35 - 45, Update
downloadFile so URL.revokeObjectURL runs in a later task after a.click(), rather
than synchronously in the same task; preserve the existing blob URL creation,
anchor download setup, click, and cleanup flow.
```

</details>

<!-- cr-comment:v1:efc24d0f1b888b39d518ab8a -->

</blockquote></details>
<details>
<summary>src/frontend/components/ReconnectingBanner.tsx-19-23 (1)</summary><blockquote>

`19-23`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Add a polite live region to `ReconnectingBanner`.**

`Alert` does not create a live region by default. Add `role="status"` and `aria-live="polite"` so assistive technology announces reconnect updates.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ReconnectingBanner.tsx` around lines 19 - 23, Add
role="status" and aria-live="polite" to the Alert element returned by
ReconnectingBanner, preserving the existing warning variant, title, and message
rendering.
```

</details>

<!-- cr-comment:v1:eac32fc662fb1515f35eb07d -->

</blockquote></details>
<details>
<summary>src/frontend/services/feedback-api.ts-37-42 (1)</summary><blockquote>

`37-42`: _🔒 Security & Privacy_ | _🟡 Minor_ | _⚡ Quick win_

**Keep `user_id` out of the query string.** The proxy logs the complete `agentUrl`, including `user_id`. Pass the identifier in a header or redact it before logging the URL.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/services/feedback-api.ts` around lines 37 - 42, Update the
feedback request in the fetch call to remove user_id from the query string; pass
the identifier through an appropriate request header instead, while preserving
the encoded threadId URL and existing credentials behavior.
```

</details>

<!-- cr-comment:v1:931fb1c580c7ccdd7dc92588 -->

</blockquote></details>
<details>
<summary>src/frontend/components/ErrorRecovery.tsx-133-144 (1)</summary><blockquote>

`133-144`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Drop the `aria-label` that hides the visible retry label.**

`aria-label="Retry operation"` replaces the accessible name. When `retryLabel` is `Retry (attempt 2/3)`, screen readers and voice control still get "Retry operation", so the attempt counter is never announced and the spoken name does not match the visible text. The visible text alone is a sufficient accessible name.

<details>
<summary>♿ Proposed fix</summary>

```diff
           <Button
             variant={retryVariant}
             onClick={onRetry}
             isDisabled={maxRetriesReached || isRetrying}
             isLoading={isRetrying}
             spinnerAriaValueText="Retrying"
-            aria-label="Retry operation"
           >
             {retryLabel}
           </Button>
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ErrorRecovery.tsx` around lines 133 - 144, Remove the
aria-label prop from the retry Button in ErrorRecovery’s onRetry rendering so
its accessible name comes from the visible retryLabel, preserving
attempt-specific text for screen readers and voice control.
```

</details>

<!-- cr-comment:v1:cb6a8f5b6fcc18f8811e3eff -->

</blockquote></details>
<details>
<summary>src/frontend/components/layout/AppLayout.tsx-361-378 (1)</summary><blockquote>

`361-378`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Use a `main` landmark for the skip-link target.**

`global.css` already makes `.skip-to-main` visible on focus. Do not add `focus:not-sr-only`. Replace the `div#main-content` with `main#main-content`.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/layout/AppLayout.tsx` around lines 361 - 378, Replace
the `div` with id `main-content` inside `AppLayout`’s `ErrorBoundary` with a
`main` element, preserving its existing classes and children so the skip link
targets a semantic main landmark. Do not add `focus:not-sr-only` to the
skip-link anchor.
```

</details>

<!-- cr-comment:v1:43d9c441e9925c60b78cec5c -->

</blockquote></details>
<details>
<summary>src/frontend/global.css-198-205 (1)</summary><blockquote>

`198-205`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Resolve the Stylelint keyframe naming errors.**

Stylelint reports `keyframes-name-pattern` errors for `slideInRight` (line 198) and `pulseGlow` (line 202). The names use camelCase, and the rule expects kebab-case. The pre-existing keyframes in this file use camelCase too, so pick one resolution and apply it consistently:

- Rename the keyframes to `slide-in-right` and `pulse-glow`, and update the class definitions at lines 219-220. Rename the pre-existing camelCase keyframes as well.
- Or relax `keyframes-name-pattern` in the Stylelint configuration.

Otherwise the lint job fails.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/global.css` around lines 198 - 205, Resolve the
keyframes-name-pattern violations by standardizing all keyframe identifiers in
global.css to kebab-case, including slideInRight and pulseGlow and the
pre-existing camelCase keyframes. Update every animation or animation-name
reference in the related class definitions and elsewhere in the file to use the
renamed identifiers.
```

</details>

<!-- cr-comment:v1:a352c4ce57242f1003daeb0c -->

_Source: Linters/SAST tools_

</blockquote></details>
<details>
<summary>src/frontend/global.css-171-177 (1)</summary><blockquote>

`171-177`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Include the remaining input types in the focus rule.**

The reset restoration at lines 151-156 covers `text`, `search`, `email`, `password`, and `url` inputs. The focus rule at lines 171-173 covers only `textarea`, `text`, and `search`. Email, password, and URL inputs therefore keep the unfocused border and box shadow.

<details>
<summary>🎨 Proposed fix</summary>

```diff
 textarea:focus,
 input[type="text"]:focus,
-input[type="search"]:focus {
+input[type="search"]:focus,
+input[type="email"]:focus,
+input[type="password"]:focus,
+input[type="url"]:focus {
   border-color: var(--primary);
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/global.css` around lines 171 - 177, Extend the focus selector
list in the global focus rule to include input[type="email"],
input[type="password"], and input[type="url"], preserving the existing
border-color, box-shadow, and outline declarations.
```

</details>

<!-- cr-comment:v1:a9a8ff26430cd77e5b18ec1f -->

</blockquote></details>
<details>
<summary>src/components/AnnouncementBanner.tsx-60-65 (1)</summary><blockquote>

`60-65`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Guard the `sessionStorage.setItem` call.**

`sessionStorage.setItem` throws when storage is unavailable or the quota is exceeded. The exception escapes the click handler, so `setDismissed(true)` never runs and the banner stays visible. The read path at line 47 is already inside a `try` block. Apply the same protection here.

<details>
<summary>🛡️ Proposed fix</summary>

```diff
   const handleClose = useCallback(() => {
-    if (payload?.message) {
-      sessionStorage.setItem(STORAGE_PREFIX + messageHash(payload.message), '1');
-    }
+    if (payload?.message) {
+      try {
+        sessionStorage.setItem(STORAGE_PREFIX + messageHash(payload.message), '1');
+      } catch {
+        // storage unavailable; dismiss for this render only
+      }
+    }
     setDismissed(true);
   }, [payload?.message]);
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/components/AnnouncementBanner.tsx` around lines 60 - 65, Wrap the
sessionStorage.setItem call in handleClose with try/catch so storage failures do
not escape the click handler; keep setDismissed(true) outside the guarded
operation so the banner is dismissed whether storage is available or not.
```

</details>

<!-- cr-comment:v1:a47895c3657c32ee92856eb1 -->

</blockquote></details>
<details>
<summary>index.html-9-20 (1)</summary><blockquote>

`9-20`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Default to dark when `theme` is missing from stored settings.**

If `template-ui-settings` contains `{}`, `JSON.parse(stored).theme` is `undefined`. The script removes the dark classes, while the application defaults to dark. Use a dark fallback for missing `theme`.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@index.html` around lines 9 - 20, Update the theme initialization around
localStorage key template-ui-settings so a missing or undefined parsed theme
defaults to 'dark', preserving explicitly stored non-dark themes and the
existing catch fallback.
```

</details>

<!-- cr-comment:v1:9cf7f57ad468c7b8f1031037 -->

</blockquote></details>
<details>
<summary>src/frontend/hooks/useRateLimitState.ts-46-59 (1)</summary><blockquote>

`46-59`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Derive the countdown from `resetTime`, and move `clearTick` out of the state updater.**

Two problems exist in this interval callback:

1. The countdown decrements a counter once per tick. Browsers throttle timers in background tabs, so ticks are dropped. `retryAfterSeconds` then drifts above the true remaining time, and the UI reports a rate limit after the limit expired. Compute the remaining seconds from `resetTime` instead.
2. `clearTick()` runs inside the `setState` updater at line 50. React may invoke an updater more than once, so the side effect can repeat. `clearInterval` is idempotent, so the current impact is limited, but the pattern is unsafe.

Both problems disappear if the tick computes the remaining time from the stored `resetTime` and clears the interval from an effect.

<details>
<summary>🛠️ Proposed change</summary>

```diff
-      tickRef.current = setInterval(() => {
-        setState((prev) => {
-          const next = prev.retryAfterSeconds - 1;
-          if (next <= 0) {
-            clearTick();
-            return { isRateLimited: false, retryAfterSeconds: 0, resetTime: null };
-          }
-          return {
-            isRateLimited: true,
-            retryAfterSeconds: next,
-            resetTime: prev.resetTime,
-          };
-        });
-      }, 1000);
+      tickRef.current = setInterval(() => {
+        const remaining = Math.ceil((resetTime.getTime() - Date.now()) / 1000);
+        if (remaining <= 0) {
+          clearTick();
+          setState({ isRateLimited: false, retryAfterSeconds: 0, resetTime: null });
+          return;
+        }
+        setState({ isRateLimited: true, retryAfterSeconds: remaining, resetTime });
+      }, 1000);
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/hooks/useRateLimitState.ts` around lines 46 - 59, Update the
interval callback in the rate-limit state hook to derive remaining seconds from
the stored resetTime and the current time, rather than decrementing
prev.retryAfterSeconds, so throttled timers cannot cause drift. Keep the state
updater pure by removing clearTick() from it, and move interval cleanup to an
effect that clears the timer once the computed remaining time reaches zero.
```

</details>

<!-- cr-comment:v1:eb49fd994ec7a3a4f60ab216 -->

_Source: Linters/SAST tools_

</blockquote></details>
<details>
<summary>package.json-80-95 (1)</summary><blockquote>

`80-95`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Raise the minimum supported Node.js version to 22.13.0.**

`jsdom@29.1.1` requires Node.js `^20.19.0 || ^22.13.0 || >=24.0.0`, but the project documents Node.js `>=22.0.0`.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@package.json` around lines 80 - 95, Update the project’s documented and
enforced minimum Node.js version from 22.0.0 to 22.13.0, including the relevant
package.json engines or version metadata and any repository configuration that
enforces it. Keep the existing Node.js major-version support unchanged while
ensuring all declarations consistently require at least 22.13.0.
```

</details>

<!-- cr-comment:v1:bae5793d4453170ab1d9abb3 -->

</blockquote></details>
<details>
<summary>src/components/AnnouncementBanner.tsx-74-74 (1)</summary><blockquote>

`74-74`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Use the project border token.**

`--pf-v5-global--BorderColor--200` is not available with PatternFly 6. Use the defined `--border` token instead.

<details>
<summary>🎨 Proposed fix</summary>

```diff
-    <div className="w-full shrink-0 border-b border-[var(--pf-v5-global--BorderColor--200)]">
+    <div className="w-full shrink-0 border-b border-border">
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/components/AnnouncementBanner.tsx` at line 74, Update the border class on
the outer div in AnnouncementBanner to use the project’s --border token instead
of the unavailable --pf-v5-global--BorderColor--200 token, preserving the
existing border-b styling.
```

</details>

<!-- cr-comment:v1:72eed07b9067b83d66d10fc4 -->

</blockquote></details>
<details>
<summary>src/frontend/services/authenticated-fetch.ts-59-59 (1)</summary><blockquote>

`59-59`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Use `same-origin` credentials and add a client-side timeout.**

The BFF already aborts upstream agent requests after 30 seconds and returns `502`. A client-side timeout still bounds browser-to-BFF stalls. Compose it with `init.signal`, including an already-aborted signal.

All `agent-rest` URLs use the same-origin `/api/proxy/agent` BFF. Replace `credentials: 'include'` with `credentials: 'same-origin'`.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/services/authenticated-fetch.ts` at line 59, Update the fetch
logic in authenticated-fetch to use credentials: 'same-origin' for the
same-origin agent-rest BFF requests. Add a client-side timeout that composes
with init.signal, correctly handling an already-aborted caller signal, and
aborts the request when the timeout expires while preserving the existing fetch
behavior.
```

</details>

<!-- cr-comment:v1:330c8c818d6cb0a247950bda -->

</blockquote></details>
<details>
<summary>src/frontend/components/ActivityTimeline.tsx-61-61 (1)</summary><blockquote>

`61-61`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Tailwind v3 leading `!` important modifiers remain after the upgrade to Tailwind 4.1.5.** In Tailwind v4 the important modifier is a suffix, so every class written as `!utility` produces no CSS and the intended override silently does not apply.
- `src/frontend/components/ActivityTimeline.tsx#L61-L61`: change `!border-none` to `border-none!` and `!bg-neutral-700` to `bg-neutral-700!`.
- `src/frontend/components/ChatMessagesView.tsx#L397-L397`: change `[&_p]:!mb-1.5` to `[&_p]:mb-1.5!` and `[&_p:last-child]:!mb-0` to `[&_p:last-child]:mb-0!`.

Run the verification script attached to the `ActivityTimeline.tsx` comment to find any remaining occurrences across the repository.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ActivityTimeline.tsx` at line 61, Update the Tailwind
important modifiers in ActivityTimeline.tsx at lines 61-61, changing border-none
and bg-neutral-700 to suffix notation. Also update ChatMessagesView.tsx at lines
397-397, converting both paragraph margin modifiers to suffix notation. Run the
attached verification script to locate and update any remaining leading-!
Tailwind utilities across the repository.
```

</details>

<!-- cr-comment:v1:e7d92fc3dc4a7f078d6043ba -->

</blockquote></details>
<details>
<summary>src/frontend/components/ChatMessagesView.tsx-819-825 (1)</summary><blockquote>

`819-825`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Remove the duplicate live region for assistant responses.**

Line 819 declares a dedicated `aria-live="polite"` region that receives the full last assistant response. Line 860 declares the message container as `role="log" aria-live="polite"`, and the assistant message is also appended there. A screen reader announces the same response twice.

Set the message container to `aria-live="off"` and keep the single dedicated announcer, or drop the dedicated announcer and rely on the log.

<details>
<summary>♿ Proposed fix</summary>

```diff
-        <div role="log" aria-label="Chat messages" aria-live="polite" aria-busy={isLoading} className="p-4 md:p-6 space-y-5 max-w-3xl mx-auto pt-8">
+        <div role="log" aria-label="Chat messages" aria-live="off" aria-busy={isLoading} className="p-4 md:p-6 space-y-5 max-w-3xl mx-auto pt-8">
```
</details>






Also applies to: 860-860

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ChatMessagesView.tsx` around lines 819 - 825, Remove
the duplicate live-region announcement in ChatMessagesView by retaining only one
announcement mechanism: either set the message container’s `role="log"` live
behavior at the referenced message-container symbol to `aria-live="off"` while
keeping the dedicated `srAnnouncement` region, or remove the dedicated announcer
and rely on the log. Ensure each assistant response is announced exactly once.
```

</details>

<!-- cr-comment:v1:e7b781ac23169e1b60f0b32a -->

</blockquote></details>
<details>
<summary>src/frontend/components/McpStatusPanel.tsx-36-36 (1)</summary><blockquote>

`36-36`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Give the labelled container a role.**

`aria-label` on a generic `div` is ignored, because the element exposes no role. Assistive technology does not announce "MCP tool status". `TodoStrip.tsx` line 34 already pairs `role="region"` with `aria-label`; use the same pattern here.

<details>
<summary>♿ Proposed fix</summary>

```diff
-    <div className="max-w-3xl mx-auto px-3 pt-2" aria-label="MCP tool status">
+    <div className="max-w-3xl mx-auto px-3 pt-2" role="region" aria-label="MCP tool status">
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/McpStatusPanel.tsx` at line 36, Update the labelled
container in McpStatusPanel to include role="region" alongside its existing
aria-label, matching the established TodoStrip pattern so assistive technology
announces “MCP tool status.”
```

</details>

<!-- cr-comment:v1:110b1ba78fec18fb21d1db38 -->

</blockquote></details>
<details>
<summary>src/frontend/lib/streaming/StreamingManager.ts-146-165 (1)</summary><blockquote>

`146-165`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Add a connection timeout to the stream request.**

The `fetch` call has no timeout. If the proxy accepts the connection and then stalls before the first byte, `stream()` stays in `'connecting'` forever and no callback fires. Add a timer that aborts the controller if the response headers do not arrive, and clear it once the reader is obtained.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/lib/streaming/StreamingManager.ts` around lines 146 - 165, Add a
connection-timeout timer around the fetch call in the stream() flow, aborting
the request controller if response headers do not arrive before the timeout.
Clear the timer immediately after response.body?.getReader() successfully
returns in the existing reader acquisition path, while preserving current HTTP
error and missing-reader handling.
```

</details>

<!-- cr-comment:v1:19792151ac54ce1e10578767 -->

</blockquote></details>
<details>
<summary>src/frontend/components/ArtifactViewer.tsx-43-43 (1)</summary><blockquote>

`43-43`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Tailwind v4 important modifier is a suffix, not a prefix.** Both files use the v3 `!utility` form. The project declares Tailwind CSS 4.1.5, where the modifier must follow the utility name, so neither override is generated.
- `src/frontend/components/ArtifactViewer.tsx#L43-L43`: change `className="!p-1"` to `className="p-1!"`.
- `src/frontend/components/TasksSidebar.tsx#L81-L81`: change `!pt-0` to `pt-0!` in the `CardBody` class list.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ArtifactViewer.tsx` at line 43, Update the Tailwind
important modifiers in ArtifactViewer.tsx:43-43 and TasksSidebar.tsx:81-81 to
Tailwind v4 suffix syntax, changing the padding utilities to p-1! and pt-0!
respectively while preserving the surrounding class lists.
```

</details>

<!-- cr-comment:v1:aade95db95b19374e9df3091 -->

</blockquote></details>
<details>
<summary>src/frontend/components/ArtifactViewer.tsx-26-30 (1)</summary><blockquote>

`26-30`: _🩺 Stability & Availability_ | _🟡 Minor_ | _⚡ Quick win_

**Handle clipboard failures.**

`navigator.clipboard` is undefined in non-secure contexts, and `writeText` rejects when the permission is denied. The rejection is unhandled here, so the copy silently fails and `copied` never changes.

<details>
<summary>🛡️ Proposed fix</summary>

```diff
   const handleCopy = async () => {
-    await navigator.clipboard.writeText(content);
-    setCopied(true);
-    setTimeout(() => setCopied(false), 2000);
+    try {
+      await navigator.clipboard?.writeText(content);
+      setCopied(true);
+      setTimeout(() => setCopied(false), 2000);
+    } catch {
+      // clipboard unavailable or permission denied
+    }
   };
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ArtifactViewer.tsx` around lines 26 - 30, Update
handleCopy to guard against navigator.clipboard being unavailable and catch
failures from writeText, ensuring rejected or unsupported clipboard operations
do not remain unhandled and copied is only set after a successful write.
Preserve the existing success reset behavior.
```

</details>

<!-- cr-comment:v1:ec493e8c42003039e43508fb -->

</blockquote></details>
<details>
<summary>src/frontend/types/user.ts-22-23 (1)</summary><blockquote>

`22-23`: _🗄️ Data Integrity & Integration_ | _🟡 Minor_ | _⚡ Quick win_

**Fix development `AppData` injection.**

The server payload and test setup provide `agentName` and `basePath`, but `index.html` provides neither. `main.tsx` then replaces the inline `window.APP_DATA` with `{}` because `#root` has no `data-app`. Preserve the inline payload or add a complete `data-app` payload with the required fields.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/types/user.ts` around lines 22 - 23, Update the frontend
bootstrap flow around main.tsx and AppData so index.html does not overwrite the
inline window.APP_DATA with an empty object when `#root` lacks data-app. Preserve
the existing inline payload, or provide a complete data-app value containing the
required agentName and basePath fields, while keeping server and test payload
handling intact.
```

</details>

<!-- cr-comment:v1:9553dbdfe045873c909e65f3 -->

</blockquote></details>
<details>
<summary>src/frontend/utils/errorHandler.ts-66-72 (1)</summary><blockquote>

`66-72`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Handle Safari and plain abort errors**

`fromFetchError` returns `UNKNOWN_ERROR` for Safari’s `TypeError: Load failed` and for a plain `Error` with `name === 'AbortError'`. Check the abort name on `Error` instances first, then extend the network-message check to include `Load failed`. Do not classify every `TypeError`, because generic `TypeError` values can indicate programming errors. This helper currently has no callers, so the retry-path impact is not present.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/utils/errorHandler.ts` around lines 66 - 72, Update
ErrorHandler.fromFetchError to check Error instances with name === 'AbortError'
before the existing DOMException handling, returning STREAM_INTERRUPTED for
those cancellations. Extend the network-error message check to recognize
Safari’s TypeError message “Load failed” while retaining the existing fetch
check, and do not classify unrelated TypeError instances as network errors.
```

</details>

<!-- cr-comment:v1:b8abf271d02412c7adff8248 -->

</blockquote></details>
<details>
<summary>src/frontend/lib/streaming/SSEProcessor.ts-142-147 (1)</summary><blockquote>

`142-147`: _🗄️ Data Integrity & Integration_ | _🟡 Minor_ | _⚡ Quick win_

**Bound incomplete SSE input.**

`feed` retains input without a `\n\n` delimiter indefinitely. Add a maximum buffer size and reset or report an error when the limit is exceeded. The proxy emits LF delimiters, so CRLF normalization is not required in this frontend path.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/lib/streaming/SSEProcessor.ts` around lines 142 - 147, Update
SSEProcessor.feed to enforce a maximum size for the retained incomplete
this.buffer after splitting on LF delimiters. When the buffer exceeds that limit
without a complete \n\n event boundary, reset it or report an error using the
class’s existing error-handling approach; preserve normal parsing for valid
input and do not add CRLF normalization.
```

</details>

<!-- cr-comment:v1:8a592e453d3bc0128e6f772e -->

</blockquote></details>
<details>
<summary>src/frontend/pages/HomePage.tsx-8-13 (1)</summary><blockquote>

`8-13`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Move `QUICK_PROMPTS` into `HomePage`.** The server injects `agentName` into `root[data-app]`, but `main.tsx` assigns that data to `window.APP_DATA` only after imported modules evaluate. `QUICK_PROMPTS` can therefore capture `'Agent'`, while render-time reads use the configured name. Derive `agentName` once inside `HomePage` and build `quickPrompts` from it.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/pages/HomePage.tsx` around lines 8 - 13, Move the QUICK_PROMPTS
definition into the HomePage component, derive agentName once from
window.APP_DATA there, and build the component-local quickPrompts from that
value so it reflects the server-injected name at render time.
```

</details>

<!-- cr-comment:v1:5e077e2147ca64be89657aaf -->

</blockquote></details>
<details>
<summary>src/frontend/test-utils/setup.ts-90-98 (1)</summary><blockquote>

`90-98`: _📐 Maintainability & Code Quality_ | _🟡 Minor_ | _⚡ Quick win_

**Add `preferred_username` to the `window.USER_DATA` fixture.**

Application code reads `window.USER_DATA.preferred_username`, for example in `src/frontend/contexts/ChatContext.tsx` line 102 and in `AppLayout`. The fixture omits that field, so those paths receive `undefined` in tests and can hide real failures.




<details>
<summary>🐛 Proposed fix</summary>

```diff
 Object.defineProperty(window, 'USER_DATA', {
   value: {
     displayName: 'Test User',
     given_name: 'Test',
+    preferred_username: 'test.user',
     email: 'test@example.com',
   },
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/test-utils/setup.ts` around lines 90 - 98, Update the
window.USER_DATA fixture in the test setup to include a representative
preferred_username value alongside the existing user fields, so ChatContext and
AppLayout tests receive the same property as application code.
```

</details>

<!-- cr-comment:v1:1b6922e7c5e8594d3bca6e5b -->

</blockquote></details>
<details>
<summary>src/frontend/components/Sidebar.tsx-145-162 (1)</summary><blockquote>

`145-162`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Reject empty rename input.**

`handleSaveRename` forwards `editTitle` unchanged. If the user clears the input and presses Enter or blurs the field, the chat title becomes an empty string and the list entry renders blank. Trim the value and keep the previous title when the result is empty.




<details>
<summary>🐛 Proposed fix</summary>

```diff
   const handleSaveRename = (chatId: string) => {
-    onRenameChat(chatId, editTitle);
+    const next = editTitle.trim();
+    if (next) onRenameChat(chatId, next);
     setEditingChat(null);
     setEditTitle('');
   };
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/Sidebar.tsx` around lines 145 - 162, Update
handleSaveRename in Sidebar so it trims editTitle before saving and preserves
the chat’s existing title when the trimmed result is empty. Ensure both Enter
and blur paths use this validation, while non-empty renamed titles are saved
normally.
```

</details>

<!-- cr-comment:v1:d163a2a88c4013fa87abf053 -->

</blockquote></details>
<details>
<summary>src/frontend/components/Sidebar.tsx-86-109 (1)</summary><blockquote>

`86-109`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Reset `searchQuery` when the search input is hidden.**

The search input renders only when `chatHistory.length > 3`. If the user types a query and then deletes chats until 3 or fewer remain, the input unmounts but `searchQuery` stays in state. `filteredChats` then keeps filtering, and the user sees "No matching threads" with no control to clear the filter.




<details>
<summary>🐛 Proposed fix</summary>

```diff
   const filteredChats = useMemo(() => {
-    const q = searchQuery.trim().toLowerCase();
+    const q = chatHistory.length > 3 ? searchQuery.trim().toLowerCase() : '';
     if (!q) return chatHistory;
```
</details>

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/Sidebar.tsx` around lines 86 - 109, Reset searchQuery
whenever chatHistory.length is 3 or fewer, so hiding the SearchInput also clears
the active filter. Update the state/effect logic associated with filteredChats
or the chat history in Sidebar, while preserving search behavior when more than
three chats remain.
```

</details>

<!-- cr-comment:v1:ae740411982483ccaeee95b6 -->

</blockquote></details>
<details>
<summary>src/frontend/services/logout.ts-9-16 (1)</summary><blockquote>

`9-16`: _🎯 Functional Correctness_ | _🟡 Minor_ | _⚡ Quick win_

**Prefix client auth URLs with `buildAppPath`.** With a non-root `APP_DATA.basePath`, `/auth/logout` and `/auth/login` target the origin root instead of the application route. Apply the same fix to `SessionExpiredModal` and `/auth/refresh` in `useRefreshableToken`.

<details>
<summary>🤖 Prompt for AI Agents</summary>

```
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/services/logout.ts` around lines 9 - 16, Prefix the
`/auth/login` URL in `buildGatewayLoginUrl` and the `/auth/logout` request in
`logout` with `buildAppPath`. Apply the same `buildAppPath` wrapping to the
login/logout-related URLs in `SessionExpiredModal` and the `/auth/refresh`
request in `useRefreshableToken`, preserving their existing query and request
behavior.
```

</details>

<!-- cr-comment:v1:776c1e4f1d6ed5c65f08b203 -->

</blockquote></details>

</blockquote></details>

<!-- This is an auto-generated comment by CodeRabbit for review status -->

[coderabbitai]

🎯 Functional Correctness | 🔴 Critical | ⚡ Quick win

Fix the mismatched thinking-tag regex.

Line 52 opens with <think> and closes with </redacted_thinking>. That tag pair does not exist. Two defects follow:

• <think>…</think> blocks are never extracted, so raw reasoning text renders inside the assistant Markdown body.
• If a <think> open tag and a later </redacted_thinking> close tag both appear in one message, the pattern spans both and deletes all answer text between them.

🐛 Proposed fix

   const patterns: RegExp[] = [
-    /<think>([\s\S]*?)<\/redacted_thinking>/gi,
+    /<think>([\s\S]*?)<\/think>/gi,
     /<thinking>([\s\S]*?)<\/thinking>/gi,
+    /<redacted_thinking>([\s\S]*?)<\/redacted_thinking>/gi,
   ];

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	const patterns: RegExp[] = [
	/<think>([\s\S]*?)<\/redacted_thinking>/gi,
	/<thinking>([\s\S]*?)<\/thinking>/gi,
	];
	const patterns: RegExp[] = [
	/<think>([\s\S]*?)<\/think>/gi,
	/<thinking>([\s\S]*?)<\/thinking>/gi,
	/<redacted_thinking>([\s\S]*?)<\/redacted_thinking>/gi,
	];

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/components/ChatMessagesView.tsx` around lines 51 - 54, Fix the
first pattern in the patterns array used by ChatMessagesView so it matches a
<think> opening tag with a corresponding </think> closing tag. Preserve the
existing separate <thinking>…</thinking> pattern and ensure the extractor does
not span or remove answer text between mismatched tags.

[coderabbitai]

🩺 Stability & Availability | 🔴 Critical | ⚡ Quick win

Stop the refresh interval after a 401.

The return on line 58 skips the token update but does not stop the timer. expiresAt is only reassigned on a successful refresh, so after a 401 it stays in the past and line 52 stays true. The interval on line 48 keeps firing every second. Each tick issues another /auth/refresh?forceRefresh=true request and calls notifySessionExpired() again.

Two consequences follow: one request per second per open tab against the auth endpoint for as long as the tab stays open, and repeated session-expired notifications that re-trigger the registered onAuthExpired handler.

Clear the interval when the session is unauthorized.

🐛 Proposed fix

           const response = await fetch("/auth/refresh?forceRefresh=true");
           if (response.status === 401) {
+            if (keepTokenRefreshedIntervalRef.current) {
+              clearInterval(keepTokenRefreshedIntervalRef.current);
+              keepTokenRefreshedIntervalRef.current = null;
+            }
             notifySessionExpired();
             return;
           }

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	if (response.status === 401) {
	notifySessionExpired();
	return;
	}
	if (response.status === 401) {
	if (keepTokenRefreshedIntervalRef.current) {
	clearInterval(keepTokenRefreshedIntervalRef.current);
	keepTokenRefreshedIntervalRef.current = null;
	}
	notifySessionExpired();
	return;
	}

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/frontend/hooks/useRefreshableToken.tsx` around lines 56 - 59, Update the
401 branch in the token refresh interval callback within useRefreshableToken to
clear the active refresh timer before calling notifySessionExpired and
returning. Reuse the interval handle established by the hook so unauthorized
sessions stop scheduling further refresh requests and notifications.

[coderabbitai]

Actionable comments posted: 9

🧹 Nitpick comments (2)

docs/deployment-patterns.md (1)

113-118: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick win

Add language identifiers to the fenced code blocks.

markdownlint reports MD040 for these fences. Use text for precedence and log-output examples.

Also applies to: 247-252, 351-353, 372-374, 392-395

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/deployment-patterns.md` around lines 113 - 118, Update the fenced code
blocks in the deployment patterns documentation, including the precedence and
log-output examples, to specify the text language identifier after each opening
fence. Apply this consistently to all referenced blocks so markdownlint MD040
passes.

Source: Linters/SAST tools

config/compliance/README.md (1)

28-39: 🔒 Security & Privacy | 🔵 Trivial | ⚡ Quick win

Pin and verify the OPA binary.

downloads/latest installs a mutable version without checksum or signature verification. Pin a tested OPA version and verify its published digest before installing it. This makes policy compilation reproducible and reduces supply-chain risk.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@config/compliance/README.md` around lines 28 - 39, Update the OPA
installation instructions in the compliance README to use a specific tested
version instead of the mutable downloads/latest URL, and add verification of the
vendor-published checksum or signature before making the binary executable or
using opa version. Apply the pinned, verified installation flow consistently to
the documented platform commands.

🤖 Prompt for all review comments with AI agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@config/compliance/README.md`:
- Around line 72-75: Update the fenced code examples in the compliance README,
including the OPA output, input-field list, and file-structure blocks, to
specify the text language tag. Preserve their contents while ensuring every
referenced fence passes Markdownlint MD040.
- Around line 151-171: Update the sample OPA eval command in the local testing
section to load the complete config/compliance directory instead of only
policy.rego. Replace the --data target while preserving the existing input
payload and deny query.

In `@config/ui/examples/minimal.yaml`:
- Around line 1-2: Update the header comments in the minimal configuration
example to document that both FEATURE_AUTH_ENABLED and AUTH_ENABLED must be
unset or false, while preserving the local-development/auth-disabled guidance
and noting that environment variables override YAML.

In `@config/ui/examples/production.yaml`:
- Around line 8-10: Update the configuration validation flow that evaluates OPA
and resolves agent endpoints so AGENT_HOST is parsed into a hostname before the
suffix check, rather than validating the raw URL; ensure the resolved hostname
is used for policy evaluation while preserving AGENT_ENDPOINT’s assignment to
agent.endpoint and correctly handling ports and paths.

In `@docs/deployment-patterns.md`:
- Around line 191-196: Remove the agent.endpoint entry from the
environment-driven Kind configuration example, leaving the timeout and
AGENT_HOST guidance intact so the environment variable is actually used
according to the priority order.
- Around line 44-48: Update the restart step in the deployment instructions to
match the documented hot-reload behavior: instruct users to save the title
change and refresh the browser instead of running npm start, unless the
documentation explicitly limits restarts to a separately defined condition.
- Around line 217-225: Update the Agent A and Agent B shell examples in the
deployment documentation, including the corresponding example at the other
referenced section, so BRANDING_TITLE and AGENT_ENDPOINT are exported or
assigned directly before the server command. Ensure the variables are available
to the child server process while preserving the existing values.
- Around line 108-118: Update the features.auth_enabled documentation to
describe environment variables as overrides rather than the only configuration
source. Revise the table wording to acknowledge settings.yaml, and ensure the
precedence section and related local/Kind examples consistently show environment
values taking priority over YAML.
- Around line 120-124: Update the authentication-disabled documentation around
auth_enabled to state that COOKIE_SIGN remains required and must contain at
least 32 characters, while SSO_CLIENT_ID and SSO_CLIENT_SECRET remain optional;
align the local configuration guidance accordingly.

---

Nitpick comments:
In `@config/compliance/README.md`:
- Around line 28-39: Update the OPA installation instructions in the compliance
README to use a specific tested version instead of the mutable downloads/latest
URL, and add verification of the vendor-published checksum or signature before
making the binary executable or using opa version. Apply the pinned, verified
installation flow consistently to the documented platform commands.

In `@docs/deployment-patterns.md`:
- Around line 113-118: Update the fenced code blocks in the deployment patterns
documentation, including the precedence and log-output examples, to specify the
text language identifier after each opening fence. Apply this consistently to
all referenced blocks so markdownlint MD040 passes.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Enterprise

Run ID: a7c5626b-426f-4539-a7ff-0973df54d080

📥 Commits

Reviewing files that changed from the base of the PR and between 4e0ed34 and 99e5a30.

📒 Files selected for processing (8)

• README.md
• config/compliance/README.md
• config/ui/README.md
• config/ui/examples/blue-theme.yaml
• config/ui/examples/minimal.yaml
• config/ui/examples/production.yaml
• config/ui/examples/red-hat-branding.yaml
• docs/deployment-patterns.md

[coderabbitai]

📐 Maintainability & Code Quality | 🟡 Minor | ⚡ Quick win

Add language tags to the fenced examples.

Markdownlint reports bare fences for the OPA output, input-field list, and file-structure examples. Use text for these blocks so MD040 passes.

Also applies to: 142-149, 178-183

🧰 Tools

🪛 markdownlint-cli2 (0.23.1)

[warning] 72-72: Fenced code blocks should have a language specified

(MD040, fenced-code-language)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@config/compliance/README.md` around lines 72 - 75, Update the fenced code
examples in the compliance README, including the OPA output, input-field list,
and file-structure blocks, to specify the text language tag. Preserve their
contents while ensuring every referenced fence passes Markdownlint MD040.

Source: Linters/SAST tools

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

🧩 Analysis chain

🏁 Script executed:

sed -n '120,190p' config/compliance/README.md
printf '\n--- compliance files ---\n'
git ls-files config/compliance
printf '\n--- policy references ---\n'
rg -n --glob '!*dist*' 'data\.json|opa eval|policy\.rego|config/compliance' config/compliance .github 2>/dev/null | head -200

Repository: redhat-data-and-ai/template-ui

Length of output: 2931

---

🏁 Script executed:

python3 - <<'PY'
from pathlib import Path
for p in [Path("config/compliance/README.md"), Path("config/compliance/policy.rego"), Path("config/compliance/data.json")]:
    print(f"\n--- {p} ---")
    if not p.exists():
        print("MISSING")
        continue
    lines = p.read_text().splitlines()
    if p.name == "README.md":
        for i in range(1, len(lines)+1):
            if 1 <= i <= 220:
                print(f"{i:4}: {lines[i-1]}")
    else:
        for i, line in enumerate(lines, 1):
            print(f"{i:4}: {line}")
PY
printf '\n--- all compliance entries ---\n'
git ls-files config/compliance

Repository: redhat-data-and-ai/template-ui

Length of output: 9417

---

🏁 Script executed:

set -eu
printf '%s\n' '--- README lines 151-171 ---'
nl -ba config/compliance/README.md | sed -n '151,171p'
printf '%s\n' '--- compliance files ---'
git ls-files config/compliance
printf '%s\n' '--- data.json references ---'
rg -n 'data\.json|data\.|opa eval|--data' config/compliance

Repository: redhat-data-and-ai/template-ui

Length of output: 1046

---

🏁 Script executed:

printf '%s\n' '--- OPA runtime integration ---'
rg -n -C 4 'policy_path|compile|bundle|\.rego|\.json|opa' --glob '!config/compliance/README.md' --glob '!package-lock.json' --glob '!pnpm-lock.yaml' --glob '!yarn.lock' .
printf '%s\n' '--- OPA availability ---'
if command -v opa >/dev/null 2>&1; then
  opa version
else
  printf '%s\n' 'opa is not available in the sandbox'
fi

Repository: redhat-data-and-ai/template-ui

Length of output: 50388

---

🌐 Web query:

Open Policy Agent opa eval --data directory loads Rego and JSON files official documentation

💡 Result:

Yes, the --data flag in the Open Policy Agent (OPA) CLI is used to recursively load policy and data files [1]. When a directory is provided to the --data flag, OPA will automatically and recursively load all files with the extensions.rego,.json, and.yaml contained within that directory and its subdirectories [1][2]. Key details regarding this behavior include: - File Types: The command supports Rego policies as well as JSON and YAML data files [1][3]. - Recursive Loading: When a directory is specified, the loader traverses the directory tree to find and include all matching file types [1][4]. - Flag Usage: The --data flag can be repeated to specify multiple files or directories [1][5]. - Distinction from --bundle: While --data recursively loads files to make them available under the data namespace, the --bundle flag is used to load policy bundles, which follow specific conventions regarding directory structure and manifest files [1][2]. Official documentation confirms that without the --bundle flag, OPA defaults to this recursive loading behavior for the provided paths [1].

Citations:

• 1: https://openpolicyagent.org/docs/cli
• 2: https://v0-66-0--opa-docs.netlify.app/docs/v0.66.0/cli/
• 3: https://v1-0-1--opa-docs.netlify.app/docs/v1.0.1/cli/
• 4: https://v0-55-0--opa-docs.netlify.app/docs/v0.55.0/cli/
• 5: https://openpolicyagent.org/docs

---

Load the complete policy directory in the local OPA test.

The runtime compiles config/compliance, but the command loads only policy.rego. Use --data config/compliance to include data.json and additional .rego files.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@config/compliance/README.md` around lines 151 - 171, Update the sample OPA
eval command in the local testing section to load the complete config/compliance
directory instead of only policy.rego. Replace the --data target while
preserving the existing input payload and deny query.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Document both authentication environment variables.

FEATURE_AUTH_ENABLED is the preferred override, and environment variables take precedence over YAML. The current instructions mention only legacy AUTH_ENABLED. If FEATURE_AUTH_ENABLED=true exists in the environment, copying this file still enables SSO.

Update the comments to require both FEATURE_AUTH_ENABLED and AUTH_ENABLED to be unset or false.

Proposed documentation fix

-# Copy to config/ui/settings.yaml and ensure AUTH_ENABLED is not set (or set to false).
+# Copy to config/ui/settings.yaml and ensure FEATURE_AUTH_ENABLED and AUTH_ENABLED are not set (or set to false).
...
-  auth_enabled: false       # Bypasses SSO; uses dummy user developer@redhat.com
+  auth_enabled: false       # Bypasses SSO; keep FEATURE_AUTH_ENABLED and AUTH_ENABLED unset or false

Also applies to: 12-14

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@config/ui/examples/minimal.yaml` around lines 1 - 2, Update the header
comments in the minimal configuration example to document that both
FEATURE_AUTH_ENABLED and AUTH_ENABLED must be unset or false, while preserving
the local-development/auth-disabled guidance and noting that environment
variables override YAML.

[coderabbitai]

🩺 Stability & Availability | 🟠 Major | 🏗️ Heavy lift

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
set -euo pipefail

rg -n -C 6 'AGENT_HOST|AGENT_ENDPOINT|agent\.endpoint|internal_endpoint_suffixes|input\.agent' \
  src/server/router/proxy.router.ts \
  src/server/utils/settings.ts \
  src/server/utils/opa.ts \
  config/compliance/policy.rego

Repository: redhat-data-and-ai/template-ui

Length of output: 6932

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

printf '%s\n' '--- production config ---'
cat -n config/ui/examples/production.yaml | sed -n '1,80p'

printf '%s\n' '--- policy ---'
cat -n config/compliance/policy.rego | sed -n '1,140p'

printf '%s\n' '--- OPA call sites and settings lifecycle ---'
rg -n -C 8 'evaluate|OPA|opa|fail_on_violation|getSettings|loadSettings|validateConfig|input\.agent|agent:\s*\{' \
  src/server config --glob '!**/node_modules/**' --glob '!**/dist/**' | sed -n '1,320p'

printf '%s\n' '--- settings sections ---'
cat -n src/server/utils/settings.ts | sed -n '160,235p;280,420p;480,570p'

Repository: redhat-data-and-ai/template-ui

Length of output: 38863

---

🏁 Script executed:

#!/bin/bash
set -euo pipefail

printf '%s\n' '--- OPA plugin ---'
cat -n src/server/plugins/opa.plugin.ts | sed -n '1,260p'

printf '%s\n' '--- server startup and config loading ---'
rg -n -C 10 'setupServer|startConfigWatcher|listen\(|getSettings\(|UI_CONFIG_PATH|AGENT_HOST|AGENT_ENDPOINT' \
  src/server src --glob '*.ts' --glob '*.js' | sed -n '1,360p'

printf '%s\n' '--- settings interface and environment mappings ---'
cat -n src/server/utils/settings.ts | sed -n '1,165p;420,490p'

printf '%s\n' '--- deterministic precedence probe ---'
python3 - <<'PY'
from urllib.parse import urlsplit

suffixes = [".svc.cluster.local", ".internal.example.com"]

def opa_denies(endpoint):
    return endpoint != "" and not any(endpoint.endswith(s) for s in suffixes)

def settings_endpoint(yaml_endpoint="", agent_endpoint=None, agent_host=None):
    # Mirrors the source paths: AGENT_ENDPOINT is applied to config; AGENT_HOST
    # is consumed later by proxy/getAgentName, not applied to config.
    endpoint = yaml_endpoint
    if agent_endpoint:
        endpoint = agent_endpoint
    return endpoint

cases = [
    ("YAML empty + AGENT_HOST example", "", None, "http://my-agent-engine:8082"),
    ("YAML empty + AGENT_ENDPOINT approved URL", "", "https://agent.svc.cluster.local", None),
    ("YAML empty + AGENT_ENDPOINT approved URL with port", "", "https://agent.svc.cluster.local:8082", None),
    ("YAML empty + AGENT_ENDPOINT approved URL with path", "", "https://agent.svc.cluster.local/agent", None),
    ("YAML empty + AGENT_ENDPOINT example", "", "http://my-agent-engine:8082", None),
]
for name, yaml_endpoint, agent_endpoint, agent_host in cases:
    input_endpoint = settings_endpoint(yaml_endpoint, agent_endpoint, agent_host)
    runtime_endpoint = input_endpoint or agent_host or "http://localhost:5002"
    print({
        "case": name,
        "opa_input_agent_endpoint": input_endpoint,
        "opa_denies_endpoint_rule": opa_denies(input_endpoint),
        "runtime_endpoint": runtime_endpoint,
        "hostname": urlsplit(runtime_endpoint).hostname,
    })
PY

Repository: redhat-data-and-ai/template-ui

Length of output: 31537

---

Resolve AGENT_HOST before OPA evaluation. AGENT_ENDPOINT is applied to agent.endpoint, but AGENT_HOST is only used by the proxy. With the empty YAML value, the documented AGENT_HOST bypasses the endpoint rule and can target an unapproved host. Parse the URL hostname before applying the suffix check so ports and paths do not cause false violations.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@config/ui/examples/production.yaml` around lines 8 - 10, Update the
configuration validation flow that evaluates OPA and resolves agent endpoints so
AGENT_HOST is parsed into a hostname before the suffix check, rather than
validating the raw URL; ensure the resolved hostname is used for policy
evaluation while preserving AGENT_ENDPOINT’s assignment to agent.endpoint and
correctly handling ports and paths.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Align the restart instruction with the documented hot-reload behavior.

This step says that a title change requires a server restart. Line 70 says that branding changes, including the title, are hot-reloaded. Replace the restart instruction with save-and-refresh steps, or document the exact condition that requires a restart.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/deployment-patterns.md` around lines 44 - 48, Update the restart step in
the deployment instructions to match the documented hot-reload behavior:
instruct users to save the title change and refresh the browser instead of
running npm start, unless the documentation explicitly limits restarts to a
separately defined condition.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Describe features.auth_enabled as an environment override, not an environment-only setting.

The table says this value is set via environment variables and not YAML. The precedence section and the local and Kind examples use YAML values. Document that environment variables override YAML instead.

🧰 Tools

🪛 markdownlint-cli2 (0.23.1)

[warning] 113-113: Fenced code blocks should have a language specified

(MD040, fenced-code-language)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/deployment-patterns.md` around lines 108 - 118, Update the
features.auth_enabled documentation to describe environment variables as
overrides rather than the only configuration source. Revise the table wording to
acknowledge settings.yaml, and ensure the precedence section and related
local/Kind examples consistently show environment values taking priority over
YAML.

[coderabbitai]

🎯 Functional Correctness | 🟠 Major | ⚡ Quick win

Do not document COOKIE_SIGN as optional when authentication is disabled.

src/server/server.ts requires COOKIE_SIGN to contain at least 32 characters before it checks cfg.features.auth_enabled. Therefore, the local configuration in Lines 155-176 still needs COOKIE_SIGN, even when SSO credentials are not required.

Proposed wording

- The `COOKIE_SIGN`, `SSO_CLIENT_ID`, and `SSO_CLIENT_SECRET` env vars are not required.
+ SSO credentials are not required when authentication is disabled. `COOKIE_SIGN` is still required.

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	### When to use each flag
	**`auth_enabled: false`** — local development only. Skips the SSO plugin registration entirely. The dummy user is always `developer@redhat.com`. The `COOKIE_SIGN`, `SSO_CLIENT_ID`, and `SSO_CLIENT_SECRET` env vars are not required.
	**`debug_mode_default: true`** — useful when iterating on agent responses locally to see raw tool calls and streaming chunks. Should always be `false` in production (OPA `restrict_debug_mode: true` enforces this).
	### When to use each flag
	**`auth_enabled: false`** — local development only. Skips the SSO plugin registration entirely. The dummy user is always `developer@redhat.com`. SSO credentials are not required when authentication is disabled. `COOKIE_SIGN` is still required.
	**`debug_mode_default: true`** — useful when iterating on agent responses locally to see raw tool calls and streaming chunks. Should always be `false` in production (OPA `restrict_debug_mode: true` enforces this).

🧰 Tools

🪛 LanguageTool

[style] ~124-~124: To form a complete sentence, be sure to include a subject.
Context: ...ee raw tool calls and streaming chunks. Should always be false in production (OPA `r...

(MISSING_IT_THERE)

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/deployment-patterns.md` around lines 120 - 124, Update the
authentication-disabled documentation around auth_enabled to state that
COOKIE_SIGN remains required and must contain at least 32 characters, while
SSO_CLIENT_ID and SSO_CLIENT_SECRET remain optional; align the local
configuration guidance accordingly.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Remove the YAML endpoint from the environment-driven Kind example.

This example defines agent.endpoint and then says to use AGENT_HOST instead. The priority order in Lines 245-252 gives the YAML endpoint higher priority, so AGENT_HOST has no effect while the YAML value remains present.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/deployment-patterns.md` around lines 191 - 196, Remove the
agent.endpoint entry from the environment-driven Kind configuration example,
leaving the timeout and AGENT_HOST guidance intact so the environment variable
is actually used according to the priority order.

[coderabbitai]

🎯 Functional Correctness | 🟡 Minor | ⚡ Quick win

Export shell variables or attach them to a command.

The shell examples assign variables without export and without a command. A child process running the server will not receive these variables. Use export BRANDING_TITLE=... and export AGENT_ENDPOINT=..., or place the assignments before the server command.

Also applies to: 266-269

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@docs/deployment-patterns.md` around lines 217 - 225, Update the Agent A and
Agent B shell examples in the deployment documentation, including the
corresponding example at the other referenced section, so BRANDING_TITLE and
AGENT_ENDPOINT are exported or assigned directly before the server command.
Ensure the variables are available to the child server process while preserving
the existing values.