Software factory change

evidence/511-named-gate-diagnostics/README.md

@@ -0,0 +1,60 @@
+# Issue #511: a named gate that fails before its command now says why
+
+Every transcript here is captured output, redirected to the file by the command
+it records. Each begins with the literal invocation and ends with its exit
+status. The checkout is Linux x64; the reported incident was macOS arm64 and is
+not reproduced by any of these runs.
+
+Common preamble for every run: `RELAYFLOWD_BIN` is passed explicitly because
+`npm run test:prep` exports it inside a subshell that never reaches vitest. The
+literal value is
+`/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd` —
+the kernel target lives outside the tree because `ops/cargo.sh` redirects
+`CARGO_TARGET_DIR`. Revisions: head `f308f95`, parent `e21caad`.
+
+- `head-focused.txt` — the two new files at head: 22 passed, exit 0.
+- `mutation-reverted.txt` — mutation verification, red half.
+  `git checkout e21caad -- src/named-gate-lowering.ts` reverts the only
+  production file; blob hashes before and after the revert are recorded against
+  `git rev-parse` of both revisions. Rebuilt, re-run: 13 failed, 9 passed.
+  The 9 are stream-capture cases that already passed before the change.
+- `mutation-restored.txt` — green half. `git checkout HEAD --` restores;
+  `git hash-object` equals `git rev-parse HEAD:…`
+  (`8a4ac50391d95c95693eb20cdbc69869a5d02060`) and
+  `git status --porcelain` is empty, so the restore is byte-for-byte. Rebuilt,
+  re-run: 22 passed, exit 0.
+- `head-package-test.txt` — the required package command, `npm test` from
+  `packages/sdk`: kernel build, typecheck, build, test typecheck, then vitest.
+  Complete output including every failure. `Test Files 3 failed | 155 passed |
+  1 skipped (159)`, `Tests 30 failed | 2393 passed | 17 skipped (2440)`,
+  exit 1. The package is not green in this sandbox.
+- `head-isolated-three-files.txt` — the three failing files alone at head:
+  30 failed, 23 passed, 14 skipped.
+- `baseline-in-place-e21caad.txt` — the same three files in the same working
+  directory with `packages/sdk` reverted to `e21caad`
+  (`git diff --stat e21caad -- packages/sdk` empty, the two added test files
+  removed). Identical result: 30 failed, 23 passed, 14 skipped, same cases,
+  same line numbers. This is the baseline behind the claim that those failures
+  are pre-existing; the revision is the only variable between it and
+  `head-isolated-three-files.txt`.
+- `restore-identity.txt` — restoring the tree after that baseline.
+  `git status --porcelain -- packages/sdk` empty and all four changed blobs
+  hash-identical to `HEAD`.
+- `unrelated-failure-cause.txt` — why 22 of those failures happen here: node
+  resolves `@relayflows/surface` to two different installs, one inside the
+  checkout and one in an ancestor directory of it
+  (`/home/daytona/.relayflow-v2-supervisor/durable/node_modules`), so the
+  handle the workflow creates is absent from the `WeakMap` the test's copy of
+  `getFlowDefinition` consults (`packages/surface/src/flow.ts:137-143`). A
+  sandbox layout defect, not a code one.
+- `baseline-e21caad.txt` — the same three files at `e21caad` in a separate
+  `git worktree` under `/tmp` with `node_modules` symlinked. Reported for
+  completeness and **not** used as the baseline: only 1 failed there, because
+  `/tmp` has no ancestor `@relayflows` install, which changes the resolution
+  above. The worktree has since been removed. `baseline-in-place-e21caad.txt`
+  is the comparison that holds the environment fixed.
+
+Not run: the Rust suite (no kernel code changed) and any Cloud runner-log
+publication, whose producer does not exist in this repository. Nothing here
+demonstrates the `flows logs <run-id>` acceptance item; `summary.md` records it
+as unmet.

evidence/511-named-gate-diagnostics/baseline-e21caad.txt

@@ -0,0 +1,107 @@
+# Baseline run at the parent revision e21caad, in a detached git worktree.
+$ git -C /tmp/flows-baseline-e21caad rev-parse HEAD
+e21caad1e4020d9e2675d2840301cb7eeb6f1aac
+$ git -C /tmp/flows-baseline-e21caad status --porcelain
+?? packages/sdk/node_modules
+# node_modules is a symlink to the main checkout (npm install is not rerun):
+$ ls -l /tmp/flows-baseline-e21caad/packages/sdk/node_modules
+lrwxrwxrwx 1 daytona daytona 83 Sep 20 17:41 /tmp/flows-baseline-e21caad/packages/sdk/node_modules -> /home/daytona/.relayflow-v2-supervisor/durable/repository/packages/sdk/node_modules
+
+$ RELAYFLOWD_BIN=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd npm run build && npx vitest run tests/live-kernel.test.ts tests/stuck-run-triage.test.ts tests/authored-node-runtime.test.ts
+
+> @relayflows/sdk@2.0.22 build
+> tsc && node scripts/make-cli-executable.mjs
+
+
+ RUN  v2.1.9 /tmp/flows-baseline-e21caad/packages/sdk
+
+ ❯ tests/authored-node-runtime.test.ts (14 tests | 14 skipped) 19ms
+stdout | tests/live-kernel.test.ts
+LIVE_KERNEL relayflowd=/home/daytona/.relayflows-toolchain/target/2962130851/debug/relayflowd
+LIVE_KERNEL flows=/tmp/flows-baseline-e21caad/packages/sdk/dist/cli.js
+
+ ✓ tests/stuck-run-triage.test.ts (22 tests) 3143ms
+   ✓ stuck-run-triage shell text > collects tails with no GNU timeout on PATH, as on a stock macOS 3084ms
+stdout | tests/live-kernel.test.ts > built flows CLI against live relayflowd > hn-monitor analyze-story reaches done through the real Claude analyzer CLI
+LIVE_ANALYZER ready: claude -p --model claude-haiku-4-5-20251001 round-trip OK
+
+stdout | tests/live-kernel.test.ts > surface resume after a real daemon kill > resumes a three-step run with each successful completion exactly once
+LIVE_KERNEL kill -9 pid=80459 run=01M2ZYKSAN5HW6C58HHTTVPSA8 while step=two state=Running
+
+ ❯ tests/live-kernel.test.ts (31 tests | 1 failed) 56356ms
+   ✓ built flows CLI against live relayflowd > twenty-six-step reuses 25 durable completions after editing the failed final step 1953ms
+   ✓ built flows CLI against live relayflowd > runs rung (a), parks rung (b), and keeps JSON report-shaped 2298ms
+   ✓ built flows CLI against live relayflowd > allows a deterministic run to exceed the bounded request timeout 32442ms
+   ✓ built flows CLI against live relayflowd > follows a live worker dispatch through flows run 567ms
+   ✓ built flows CLI against live relayflowd > runs an agent CLI end to end through the SDK worker 472ms
+   ✓ built flows CLI against live relayflowd > f.agent lowers to a real agent step and dispatches through a live worker 576ms
+   ✓ built flows CLI against live relayflowd > can always get a parked run to a late-attaching worker 5577ms
+   ✓ built flows CLI against live relayflowd > reports a real manual-recovery NeedsHuman state as parked 413ms
+   ✓ built flows CLI against live relayflowd > runs hn-monitor analyze-story end-to-end via a stub agent CLI (gate 2 clause 2 demo) 418ms
+   ✓ built flows CLI against live relayflowd > hn-monitor analyze-story FAILS verification when the CLI omits required schema fields 440ms
+   ✓ built flows CLI against live relayflowd > agent step preserves the CliResult wrapper as output when the CLI emits non-JSON text 460ms
+   ✓ built flows CLI against live relayflowd > AgentWorker exposes wake_context to the CLI via RELAYFLOW_WAKE_CONTEXT env var (real analyzer prerequisite) 527ms
+   ✓ built flows CLI against live relayflowd > AgentWorker leaves RELAYFLOW_WAKE_CONTEXT UNSET when the run has no wake_context (undefined-vs-null pin) 534ms
+   ✓ built flows CLI against live relayflowd > AgentWorker passes a declared model to an identified wrapper as RELAYFLOW_MODEL 634ms
+   ✓ built flows CLI against live relayflowd > AgentWorker refuses a nonconforming journal-submitted wrapper before exposing RELAYFLOW_MODEL 470ms
+   ✓ built flows CLI against live relayflowd > AgentWorker executes the raw claude adapter with its real model flag 432ms
+   ✓ built flows CLI against live relayflowd > AgentWorker executes the raw codex adapter with its real model flag 479ms
+   ✓ built flows CLI against live relayflowd > AgentWorker leaves RELAYFLOW_MODEL UNSET when the step declares no model 541ms
+   × built flows CLI against live relayflowd > hn-monitor analyze-story reaches done through the real Claude analyzer CLI 3805ms
+     → expected { …(3) } to match object { Object (gate, verdict) }
+(1 matching property omitted from actual)
+   ✓ built flows CLI against live relayflowd > preflights before journaling and names an unreachable socket 720ms
+   ✓ built flows CLI against live relayflowd > starts exactly one daemon when two runs race for one empty data dir 431ms
+   ✓ surface resume after a real daemon kill > resumes a three-step run with each successful completion exactly once 807ms
+   ✓ a relayflow can be scheduled: tick source against live relayflowd > a tick spawns a real run whose step reports the SCHEDULED instant 627ms
+
+⎯⎯⎯⎯⎯⎯ Failed Suites 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/authored-node-runtime.test.ts [ tests/authored-node-runtime.test.ts ]
+AssertionError: expected '1.3.6' to be '1.4.0' // Object.is equality
+
+Expected: "1.4.0"
+Received: "1.3.6"
+
+ ❯ tests/authored-node-runtime.test.ts:18:77
+     16| 
+     17| beforeAll(() => {
+     18|   expect(spawnSync(bun, ['--version'], { encoding: 'utf8' }).stdout.tr…
+       |                                                                             ^
+     19|   expect(existsSync(daemon), 'build the current kernel or set RELAYFLO…
+     20|   stage = mkdtempSync(join(tmpdir(), 'authored-standalone-build-'));
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯
+
+⎯⎯⎯⎯⎯⎯⎯ Failed Tests 1 ⎯⎯⎯⎯⎯⎯⎯
+
+ FAIL  tests/live-kernel.test.ts > built flows CLI against live relayflowd > hn-monitor analyze-story reaches done through the real Claude analyzer CLI
+AssertionError: expected { …(3) } to match object { Object (gate, verdict) }
+(1 matching property omitted from actual)
+
+- Expected
++ Received
+
+  Object {
+-   "gate": "json_schema",
+-   "verdict": "pass",
++   "gate": "execution",
++   "verdict": "fail",
+  }
+
+ ❯ tests/live-kernel.test.ts:1317:49
+    1315|     // the promoted output, not the test re-deriving it: the spec is
+    1316|     // the unmodified canonical one, so this record is the gate.
+    1317|     expect(stepCompleted!.payload.verification).toMatchObject({
+       |                                                 ^
+    1318|       gate: 'json_schema',
+    1319|       verdict: 'pass',
+
+⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/2]⎯
+
+ Test Files  2 failed | 1 passed (3)
+      Tests  1 failed | 52 passed | 14 skipped (67)
+   Start at  17:41:55
+   Duration  57.63s (transform 1.01s, setup 0ms, collect 2.40s, tests 59.52s, environment 0ms, prepare 114ms)
+
+EXIT=1