feat(assistant): add tester assistant kind

src/assistant.rs

@@ -16,20 +16,25 @@ pub enum AssistantStatus {
     Archived,
 }
 
-/// One worktree entry tracked by a Reviewer assistant.
+/// One worktree entry tracked by a worktree-backed assistant.
 ///
 /// `agman_created` records whether agman set up the worktree (and the local
 /// branch) itself — drives archive cleanup. Worktrees that already existed
 /// when the reviewer was created are left intact on archive.
 #[derive(Debug, Clone, Serialize, Deserialize)]
-pub struct ReviewerWorktree {
+pub struct AssistantWorktree {
     pub repo: String,
     pub branch: String,
     pub path: PathBuf,
     pub agman_created: bool,
 }
 
-/// Discriminator for the two assistant kinds. Each kind carries its own
+#[derive(Default, Serialize, Deserialize, Clone, Copy, Debug)]
+pub struct TesterCapabilities {
+    pub browser: bool,
+}
+
+/// Discriminator for the assistant kinds. Each kind carries its own
 /// kind-specific metadata; everything else (harness stamping, inbox, tmux,
 /// telegram switcher, send-message routing, poll-target enumeration) is
 /// kind-agnostic and lives on the surrounding [`AssistantMeta`].
@@ -46,7 +51,13 @@ pub enum AssistantKind {
     },
     Reviewer {
         #[serde(default)]
-        worktrees: Vec<ReviewerWorktree>,
+        worktrees: Vec<AssistantWorktree>,
+    },
+    Tester {
+        #[serde(default)]
+        worktrees: Vec<AssistantWorktree>,
+        #[serde(default)]
+        capabilities: TesterCapabilities,
     },
 }
 
@@ -172,6 +183,11 @@ impl Assistant {
         matches!(self.meta.kind, AssistantKind::Reviewer { .. })
     }
 
+    /// True if this assistant is a Tester.
+    pub fn is_tester(&self) -> bool {
+        matches!(self.meta.kind, AssistantKind::Tester { .. })
+    }
+
     /// Write meta.json to disk. Stamps `updated_at` before writing.
     pub fn save_meta(&mut self) -> Result<()> {
         self.meta.updated_at = Utc::now();

src/cli.rs

@@ -41,7 +41,7 @@ EXAMPLES:
   agman send-message chief-of-staff @./message.md")]
     SendMessage {
         /// Target: "chief-of-staff", "telegram", "researcher:<project>--<name>",
-        /// "reviewer:<project>--<name>", or a project name (for the PM)
+        /// "reviewer:<project>--<name>", "tester:<project>--<name>", or a project name (for the PM)
         target: String,
         /// Message text (can also be provided via stdin or --file)
         #[arg(allow_hyphen_values = true)]
@@ -208,17 +208,20 @@ EXAMPLES:
         file: Option<std::path::PathBuf>,
     },
 
-    /// Create an assistant (researcher or reviewer). Defaults to Chief of
+    /// Create an assistant (researcher, reviewer, or tester). Defaults to Chief of
     /// Staff-level when --project is omitted.
     #[command(after_help = "\
 EXAMPLES:
   agman create-assistant --kind researcher --name api-investigator --description \"Investigate the API latency\"
   agman create-assistant --kind reviewer --name pr-1247 --project reviews \\
     --branch galoy:fix-deposit-flow \\
     --branch lana-dashboard:fix-deposit-flow \\
-    --description \"Review the cross-repo deposit fix\"")]
+    --description \"Review the cross-repo deposit fix\"
+  agman create-assistant --kind tester --name browser-pass --project reviews \\
+    --branch galoy:fix-deposit-flow --browser \\
+    --description \"Exercise the deposit flow in browser\"")]
     CreateAssistant {
-        /// Assistant kind: researcher or reviewer
+        /// Assistant kind: researcher, reviewer, or tester
         #[arg(long, value_enum)]
         kind: AssistantKindArg,
         /// Assistant name (alphanumeric + hyphens)
@@ -230,7 +233,7 @@ EXAMPLES:
         /// Description/initial question
         #[arg(long, short, allow_hyphen_values = true)]
         description: Option<String>,
-        // --- Researcher-only flags (rejected for reviewer) ---
+        // --- Researcher-only flags (rejected for reviewer/tester) ---
         /// Repository name (researcher only — for working directory context)
         #[arg(long)]
         repo: Option<String>,
@@ -240,11 +243,14 @@ EXAMPLES:
         /// Task ID to inherit working directory from (researcher only)
         #[arg(long)]
         task: Option<String>,
-        // --- Reviewer-only flag (repeatable; rejected for researcher) ---
-        /// `<repo>:<branch>` pair to scope the reviewer to (reviewer only;
-        /// repeat to include multiple). Required for reviewers.
+        // --- Worktree-backed flags (repeatable; rejected for researcher) ---
+        /// `<repo>:<branch>` pair to scope the reviewer/tester to; repeat
+        /// to include multiple. Required for reviewers and testers.
         #[arg(long = "branch", value_name = "REPO:BRANCH")]
         branch_pair: Vec<String>,
+        /// Request browser automation tools (tester only)
+        #[arg(long, default_value_t = false)]
+        browser: bool,
     },
 
     /// List assistants
@@ -319,6 +325,30 @@ EXAMPLES:
         description: Option<String>,
     },
 
+    /// Create a tester (alias for `create-assistant --kind tester`).
+    #[command(after_help = "\
+EXAMPLES:
+  agman create-tester --name browser-pass --project reviews \\
+    --branch galoy:fix-deposit-flow --browser \\
+    --description \"Exercise the deposit flow\"")]
+    CreateTester {
+        /// Tester name (alphanumeric + hyphens)
+        #[arg(long, short)]
+        name: String,
+        /// Project name (defaults to "chief-of-staff")
+        #[arg(long)]
+        project: Option<String>,
+        /// `<repo>:<branch>` pair (repeatable, required at least once)
+        #[arg(long = "branch", value_name = "REPO:BRANCH", required = true)]
+        branch_pair: Vec<String>,
+        /// Request browser automation tools
+        #[arg(long, default_value_t = false)]
+        browser: bool,
+        /// Description
+        #[arg(long, short, allow_hyphen_values = true)]
+        description: Option<String>,
+    },
+
     /// List researchers (alias for `list-assistants --kind researcher`).
     ListResearchers {
         /// Filter by project name
@@ -357,6 +387,7 @@ EXAMPLES:
 pub enum AssistantKindArg {
     Researcher,
     Reviewer,
+    Tester,
 }
 
 #[derive(Debug, Clone, Copy, ValueEnum)]

src/config.rs

@@ -326,7 +326,7 @@ impl Config {
 
     // --- Assistant paths ---
     //
-    // Both Researchers and Reviewers share the on-disk layout under
+    // Assistants share the on-disk layout under
     // `~/.agman/assistants/<project>--<name>/`. The kind discriminator lives
     // inside `meta.json`. Tmux session names diverge by kind so a researcher
     // and reviewer with the same name+project don't collide on resume.
@@ -363,6 +363,11 @@ impl Config {
         format!("agman-reviewer-{project}--{name}")
     }
 
+    /// Tmux session name for a tester.
+    pub fn tester_tmux_session(project: &str, name: &str) -> String {
+        format!("agman-tester-{project}--{name}")
+    }
+
     // --- Project template paths ---
 
     /// Directory where project templates are stored: ~/.agman/project-templates/

src/harness/claude.rs

@@ -67,6 +67,9 @@ impl Harness for ClaudeHarness {
                 cmd.push_str(&format!(" --resume '{}'", escaped_uuid));
             }
         }
+        if ctx.capabilities.browser {
+            cmd.push_str(" --chrome");
+        }
         cmd
     }
 

src/harness/codex.rs

@@ -4,7 +4,9 @@ use anyhow::Result;
 use std::path::{Path, PathBuf};
 use std::time::{Duration, Instant};
 
-use super::{Harness, HarnessKind, LaunchContext, RegisterContext, SessionKey};
+use super::{
+    AssistantCapabilities, Harness, HarnessKind, LaunchContext, RegisterContext, SessionKey,
+};
 
 pub struct CodexHarness;
 
@@ -43,6 +45,9 @@ impl Harness for CodexHarness {
             if ctx.no_alt_screen {
                 cmd.push_str(" --no-alt-screen");
             }
+            if ctx.capabilities.browser {
+                cmd.push_str(" -c 'mcp_servers.playwright.enabled=true'");
+            }
             cmd.push_str(&format!(" -C '{}'", cwd_str));
             cmd.push_str(&format!(" resume '{}'", escaped_name));
             return cmd;
@@ -69,6 +74,9 @@ impl Harness for CodexHarness {
             cmd.push_str(" --no-alt-screen");
         }
         cmd.push_str(&format!(" -c '{}'", dev_arg_escaped));
+        if ctx.capabilities.browser {
+            cmd.push_str(" -c 'mcp_servers.playwright.enabled=true'");
+        }
         cmd
     }
 
@@ -79,6 +87,14 @@ impl Harness for CodexHarness {
         ensure_workspace_trusted_in(&trust_file, cwd)
     }
 
+    fn ensure_capabilities_configured(&self, caps: &AssistantCapabilities) -> Result<()> {
+        if caps.browser {
+            let config_toml = super::harness_home(HarnessKind::Codex).join("config.toml");
+            ensure_browser_mcp_in(&config_toml)?;
+        }
+        Ok(())
+    }
+
     /// Paste-inject `/rename <name>` post-launch and verify the entry shows
     /// up in `~/.codex/session_index.jsonl`. Self-verifying with retry: codex
     /// step 2+ relaunches faster than first launch (file watchers warm, no
@@ -380,6 +396,81 @@ pub fn ensure_workspace_trusted_in(trust_file: &Path, cwd: &Path) -> Result<()>
     write_atomically(trust_file, toml::to_string(&doc)?.as_bytes())
 }
 
+/// Ensure the Playwright MCP server is defined in codex config but disabled
+/// by default. Tester launches opt in per process with a `-c` override.
+pub fn ensure_browser_mcp_in(config_toml_path: &Path) -> Result<()> {
+    use anyhow::Context;
+
+    let mut doc: toml::Value = if config_toml_path.exists() {
+        let text = std::fs::read_to_string(config_toml_path)
+            .with_context(|| format!("read codex config file at {}", config_toml_path.display()))?;
+        if text.trim().is_empty() {
+            toml::Value::Table(toml::value::Table::new())
+        } else {
+            toml::from_str(&text).with_context(|| {
+                format!(
+                    "parse codex config file at {} as TOML",
+                    config_toml_path.display()
+                )
+            })?
+        }
+    } else {
+        toml::Value::Table(toml::value::Table::new())
+    };
+
+    let root = doc.as_table_mut().ok_or_else(|| {
+        anyhow::anyhow!(
+            "codex config file at {} is not a TOML table",
+            config_toml_path.display()
+        )
+    })?;
+
+    let mcp_servers = root
+        .entry("mcp_servers".to_string())
+        .or_insert_with(|| toml::Value::Table(toml::value::Table::new()))
+        .as_table_mut()
+        .ok_or_else(|| {
+            anyhow::anyhow!(
+                "codex config file at {} has non-table `mcp_servers`",
+                config_toml_path.display()
+            )
+        })?;
+
+    if mcp_servers.contains_key("playwright") {
+        return Ok(());
+    }
+
+    let mut playwright = toml::value::Table::new();
+    playwright.insert(
+        "command".to_string(),
+        toml::Value::String("npx".to_string()),
+    );
+    playwright.insert(
+        "args".to_string(),
+        toml::Value::Array(vec![toml::Value::String(
+            "@playwright/mcp@latest".to_string(),
+        )]),
+    );
+    playwright.insert(
+        "env_vars".to_string(),
+        toml::Value::Array(
+            [
+                "DISPLAY",
+                "WAYLAND_DISPLAY",
+                "XAUTHORITY",
+                "XDG_RUNTIME_DIR",
+            ]
+            .into_iter()
+            .map(|v| toml::Value::String(v.to_string()))
+            .collect(),
+        ),
+    );
+    playwright.insert("enabled".to_string(), toml::Value::Boolean(false));
+    mcp_servers.insert("playwright".to_string(), toml::Value::Table(playwright));
+
+    write_atomically(config_toml_path, toml::to_string(&doc)?.as_bytes())
+}
+
 /// Write `bytes` to `dest` atomically: write to `<dest>.tmp`, fsync, rename.
 /// Creates `dest`'s parent directory if missing.
 fn write_atomically(dest: &Path, bytes: &[u8]) -> Result<()> {

src/harness/mod.rs

@@ -77,6 +77,14 @@ pub fn ensure_workspace_trusted_for_test(
     }
 }
 
+/// Test-only entrypoint for codex browser MCP configuration. Production uses
+/// `CodexHarness::ensure_capabilities_configured`, which resolves the config
+/// path from `harness_home(Codex)`.
+#[doc(hidden)]
+pub fn ensure_browser_mcp_for_test(config_toml_path: &Path) -> Result<()> {
+    codex::ensure_browser_mcp_in(config_toml_path)
+}
+
 /// Identifies which harness to use. Persisted in config + per-agent stamps.
 #[derive(Debug, Clone, Copy, Default, PartialEq, Eq, Serialize, Deserialize)]
 #[serde(rename_all = "lowercase")]
@@ -170,6 +178,12 @@ pub enum SessionKey<'a> {
     Resume(&'a str),
 }
 
+/// Optional assistant capabilities requested at launch time.
+#[derive(Default, Clone, Copy, Debug)]
+pub struct AssistantCapabilities {
+    pub browser: bool,
+}
+
 /// Static input for `Harness::build_session_command`. Names follow the
 /// harness's resume / session-listing convention so the user can reattach
 /// manually from a shell (`claude --resume <id>`, `codex resume <name>`, or
@@ -197,6 +211,8 @@ pub struct LaunchContext<'a> {
     /// pane content (needed by the inbox snippet-verification loop). Claude
     /// ignores.
     pub no_alt_screen: bool,
+    /// Optional assistant capabilities. Non-assistant launches pass default.
+    pub capabilities: AssistantCapabilities,
     /// Whether this launch pins a fresh session, resumes a prior one, or
     /// neither. See `SessionKey` for the per-variant behaviour.
     pub session_key: SessionKey<'a>,
@@ -240,6 +256,12 @@ pub trait Harness: Send + Sync {
     /// usable state and `/rename` paste-injects run as shell commands).
     fn ensure_workspace_trusted(&self, cwd: &Path) -> Result<()>;
 
+    /// Ensure any requested assistant capabilities are configured before
+    /// launch. Harnesses that do not need setup use the default no-op.
+    fn ensure_capabilities_configured(&self, _caps: &AssistantCapabilities) -> Result<()> {
+        Ok(())
+    }
+
     /// Post-launch step. Called once `is_session_ready_in` reports the
     /// foreground process is no longer a shell.
     /// - Claude: no-op.