cc-workspace 5.2.1 → 5.2.3

package/CHANGELOG.md

@@ -1,5 +1,41 @@
 # Changelog
 
+## [5.2.3] — 2026-03-09
+
+### Agent Teams — correct tool mechanism
+
+The `Teammate` tool referenced in previous versions does not exist in Claude Code.
+Agent Teams actually works via `TeamCreate` + `Agent(subagent_type, team_name)` + `SendMessage`.
+
+- **Replaced all `Teammate` references with real Agent Teams tools** — team-lead,
+  e2e-validator, dispatch-feature, qa-ruthless, incident-debug now use `TeamCreate`,
+  `TeamDelete`, and `SendMessage` instead of the non-existent `Teammate` tool.
+
+- **Teammate spawn workflow documented** — dispatch-feature now includes the explicit
+  3-step workflow: `TeamCreate` → `Agent(implementer, team_name)` → `SendMessage`.
+
+- **validate-spawn-prompt hook updated** — matcher changed from `Teammate` to `Agent`,
+  with a `subagent_type` filter to only validate implementer spawns (skips Explore).
+
+- **model-routing rule updated** — routing table now documents the real mechanism.
+
+- **Auto-enable Agent Teams on install** — `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` is
+  now automatically injected into `~/.claude/settings.json` during install/update.
+
+## [5.2.2] — 2026-03-09
+
+### Agent Teams enforcement
+
+- **Removed `Task(implementer)` from agent tools** — team-lead and e2e-validator can no
+  longer spawn implementers as subagents via Task. Implementers must be spawned through
+  Agent Teams (TeamCreate + Agent with team_name).
+
+- **e2e-validator gains Agent Teams tools** — can now delegate `--fix` repairs
+  through the proper teammate protocol.
+
+- **Auto-enable Agent Teams on install** — `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` is
+  now automatically injected into `~/.claude/settings.json` during `npx cc-workspace install`.
+
 ## [5.2.1] — 2026-03-06
 
 ### Prompt quality improvements

package/bin/cli.js

@@ -195,10 +195,30 @@ function installGlobals(force) {
   // ── Migration: clean old global skills & rules from previous versions ──
   cleanLegacyGlobals();
 
+  // Ensure Agent Teams env var is in global settings
+  ensureGlobalEnv("CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS", "1");
+
   writeVersion(PKG.version);
   return true;
 }
 
+// ─── Ensure env var in global ~/.claude/settings.json ────────
+function ensureGlobalEnv(key, value) {
+  const globalSettingsPath = path.join(CLAUDE_DIR, "settings.json");
+  let settings = {};
+  try { settings = JSON.parse(fs.readFileSync(globalSettingsPath, "utf8")); }
+  catch { /* file missing or invalid — start fresh */ }
+
+  if (!settings.env) settings.env = {};
+  if (settings.env[key] === value) return; // already set
+
+  settings.env[key] = value;
+  if (DRY_RUN) { dryOk(`set ${key}=${value} in ~/.claude/settings.json`); return; }
+  mkdirp(CLAUDE_DIR);
+  writeFile(globalSettingsPath, JSON.stringify(settings, null, 2) + "\n");
+  ok(`${key} ${c.dim}(~/.claude/settings.json)${c.reset}`);
+}
+
 // ─── Clean legacy global skills & rules ─────────────────────
 // Previous versions installed skills and rules into ~/.claude/skills/
 // and ~/.claude/rules/. These now live in orchestrator/.claude/ locally.
@@ -271,7 +291,7 @@ function generateSettings(orchDir) {
         // guard-session-checkout.sh is NOT here — it's in implementer agent
         // frontmatter only. team-lead doesn't have Bash, and teammates don't
         // inherit orchestrator hooks.
-        withMatcher("Teammate", "validate-spawn-prompt.sh", 5)
+        withMatcher("Agent", "validate-spawn-prompt.sh", 5)
       ],
       SessionStart: [
         withoutMatcher("orphan-cleanup.sh", 10),
@@ -311,7 +331,7 @@ You are the tech lead. You never write application code in repos.
 You clarify, plan, manage git directly, delegate to teammates, run micro-QA.
 
 ## Security
-- \`tools\`: Read, Write, Edit, Bash, Glob, Grep, Task(implementer, Explore), Teammate, SendMessage
+- \`tools\`: Read, Write, Edit, Bash, Glob, Grep, Task(Explore), TeamCreate, TeamDelete, SendMessage
 - Bash allowed for: git (branch, worktree, log), test/typecheck in /tmp/ worktrees (micro-QA)
 - Hook \`PreToolUse\` path-aware: Write/Edit/MultiEdit only allowed in orchestrator/
 

package/global-skills/agents/e2e-validator.md

@@ -1,6 +1,6 @@
 ---
 name: e2e-validator
-prompt_version: 5.2.1
+prompt_version: 5.2.2
 description: >
   E2E validation agent for completed plans. On first boot, sets up the E2E
   environment (docker-compose, test config). On subsequent boots, validates
@@ -11,7 +11,7 @@ description: >
 model: opus
 tools: >
   Read, Write, Edit, Bash, Glob, Grep,
-  Task(implementer, Explore),
+  Task(Explore), TeamCreate, TeamDelete, SendMessage,
   mcp__chrome-devtools__navigate_page,
   mcp__chrome-devtools__click,
   mcp__chrome-devtools__fill,
@@ -40,7 +40,7 @@ maxTurns: 150
 
 ## CRITICAL — Non-negotiable rules (read FIRST)
 
-1. **NEVER modify application code** — delegate via `--fix` + Teammate(implementer)
+1. **NEVER modify application code** — delegate via `--fix` + Agent Teams (TeamCreate + Agent(implementer, team_name))
 2. **No worktrees** — work directly on repos at the correct branch (merged or session)
 3. **Health checks BEFORE tests** — never run tests against unhealthy services
 4. **Always teardown** — `docker compose down -v` even on failure
@@ -242,7 +242,7 @@ Not tied to a specific plan — runs all existing E2E tests.
 If failures exist after report and user confirms:
 
 1. Read session JSON for the plan → get `worktree_path` per repo (still exists if session not closed)
-2. Spawn one **Teammate(implementer)** per failing repo with:
+2. Create a team (`TeamCreate`), then spawn one **Agent(subagent_type: "implementer", team_name: ...)** per failing repo with:
    - `worktree_path` from session JSON (or source branch if already merged)
    - Failure details from the E2E report
    - Session branch

package/global-skills/agents/team-lead.md

@@ -1,6 +1,6 @@
 ---
 name: team-lead
-prompt_version: 5.2.1
+prompt_version: 5.2.2
 description: >
   Main orchestrator for multi-service workspaces. Clarifies specs,
   plans in markdown, manages git (branches, worktrees) directly,
@@ -9,7 +9,7 @@ description: >
   Never codes in repos — can write in orchestrator/ and run git commands.
   Triggered via claude --agent team-lead.
 model: opus
-tools: Read, Write, Edit, Bash, Glob, Grep, Task(implementer, Explore), Teammate, SendMessage
+tools: Read, Write, Edit, Bash, Glob, Grep, Task(Explore), TeamCreate, TeamDelete, SendMessage
 memory: project
 maxTurns: 200
 hooks:
@@ -135,7 +135,7 @@ This table is your quick reference — **defer to the skill for specifics**.
 | 2 — Plan | Write ./plans/{name}.md from _TEMPLATE.md | Wait for user validation |
 | 2.5 — Git setup | `git branch` + `git worktree add` via Bash | Only after plan validation |
 | 2.9 — Pre-dispatch | Verify branches + worktrees exist and are clean | Auto-fix simple cases |
-| 3 — Dispatch | ONE Teammate per repo, all commits sequential | See @dispatch-feature/references/spawn-templates.md |
+| 3 — Dispatch | ONE teammate per repo (via TeamCreate + Agent), all commits sequential | See @dispatch-feature/references/spawn-templates.md |
 | 4 — Micro-QA | Bash tests + Haiku diff after each commit | Green light or fix instruction |
 | 5 — Post-impl | cross-service → qa-ruthless → reviewer → (security-auditor if needed) → merge-prep → retro | All mandatory except security-auditor |
 

package/global-skills/dispatch-feature/SKILL.md

@@ -12,7 +12,7 @@ description: >
   "dispatch", "start dev", "launch teammates", or provides a spec.
 argument-hint: "[feature description]"
 context: fork
-allowed-tools: Read, Write, Bash, Glob, Grep, Task, Teammate, SendMessage
+allowed-tools: Read, Write, Bash, Glob, Grep, Task, TeamCreate, TeamDelete, SendMessage
 ---
 
 # Dispatch Feature — Clarify, Plan, Git, Delegate, Track
@@ -177,12 +177,23 @@ Only proceed to Phase 3 once all checks pass.
 
 ## Phase 3: Dispatch — one teammate per repo
 
-**ONE teammate per repo.** Each teammate receives the complete repo plan and handles
-all its commit units sequentially. It signals after each commit and waits for green light.
+**ONE teammate per repo.** Use `TeamCreate` to create a team for the session, then spawn
+each repo's implementer via `Agent(subagent_type: "implementer", team_name: ...)`.
+Each teammate receives the complete repo plan and handles all its commit units sequentially.
+It signals after each commit via `SendMessage` and waits for green light.
 
 This is fundamentally different from the old model (one subagent per commit unit).
 The teammate has the full picture of its repo and can escalate intelligently at any point.
 
+### Teammate spawn — workflow
+
+```
+1. TeamCreate(team_name: "session-{name}")
+2. For each repo in wave:
+   Agent(subagent_type: "implementer", team_name: "session-{name}", prompt: "...")
+3. SendMessage to coordinate (green light, fix instructions, STOP)
+```
+
 ### Teammate spawn prompt — context tiering
 
 See @references/spawn-templates.md for full templates.

package/global-skills/hooks/validate-spawn-prompt.sh

@@ -1,6 +1,6 @@
 #!/usr/bin/env bash
 # validate-spawn-prompt.sh
-# PreToolUse hook (matcher: Teammate): validates that teammate spawn prompts
+# PreToolUse hook (matcher: Agent): validates that teammate spawn prompts
 # contain the required context before allowing the spawn.
 # v5.0: Updated for one-teammate-per-repo model with worktree_path and signal protocol.
 # ALL checks are non-blocking warnings (exit 0 + stdout).
@@ -8,6 +8,12 @@ set -euo pipefail
 
 INPUT=$(cat)
 
+# Only validate implementer spawns (skip Explore and other subagent types)
+SUBAGENT_TYPE=$(echo "$INPUT" | jq -r '.tool_input.subagent_type // empty' 2>/dev/null) || true
+if [ "$SUBAGENT_TYPE" != "implementer" ]; then
+    exit 0
+fi
+
 # Extract the spawn prompt from tool input
 PROMPT=$(echo "$INPUT" | jq -r '.tool_input.prompt // empty' 2>/dev/null) || true
 
@@ -48,12 +54,12 @@ fi
 
 # Check 3: Worktree path (v5 — teammates receive a ready worktree)
 if ! echo "$PROMPT" | grep -qiE '(worktree.path|worktree.ready|go directly|/tmp/[a-z])' 2>/dev/null; then
-    BLOCKERS+="- Missing worktree_path in spawn prompt. Teammate needs the /tmp/ path prepared by Opus.\n"
+    BLOCKERS+="- Missing worktree_path in spawn prompt. Implementer needs the /tmp/ path prepared by Opus.\n"
 fi
 
 # Check 4: Signal protocol instruction
 if ! echo "$PROMPT" | grep -qiE '(SendMessage|signal protocol|commit N done|green light|wait for)' 2>/dev/null; then
-    BLOCKERS+="- Missing signal protocol instruction. Teammate must know to SendMessage after each commit and wait for green light.\n"
+    BLOCKERS+="- Missing signal protocol instruction. Implementer must know to SendMessage after each commit and wait for green light.\n"
 fi
 
 # Check 5: CLAUDE.md instruction

package/global-skills/incident-debug/SKILL.md

@@ -8,7 +8,7 @@ description: >
   stack traces or error logs. Also French triggers: "erreur", "ça marche pas".
 argument-hint: "[error description or stack trace]"
 context: fork
-allowed-tools: Read, Write, Glob, Grep, Task, Teammate, SendMessage
+allowed-tools: Read, Write, Glob, Grep, Task, TeamCreate, TeamDelete, SendMessage
 ---
 
 # Incident Debug — Multi-Layer Investigation

package/global-skills/qa-ruthless/SKILL.md

@@ -10,7 +10,7 @@ description: >
 argument-hint: "[plan-name or 'all']"
 context: fork
 model: opus
-allowed-tools: Read, Write, Glob, Grep, Task, Teammate, SendMessage
+allowed-tools: Read, Write, Glob, Grep, Task, TeamCreate, TeamDelete, SendMessage
 ---
 
 # QA Ruthless — Adversarial Quality Review

package/global-skills/rules/model-routing.md

@@ -20,9 +20,9 @@ If you write application code in a repo (not a markdown plan, not a git command,
 | Role | Model | Mechanism |
 |------|-------|-----------|
 | Orchestrator | **Opus** | `claude --agent team-lead` (frontmatter `model: opus`) |
-| Implementation teammates | **Sonnet** | `Teammate` tool via Agent Teams |
+| Implementation teammates | **Sonnet** | `TeamCreate` + `Agent(implementer, team_name)` via Agent Teams |
 | QA orchestration | **Opus** | `qa-ruthless` skill (frontmatter `model: opus`) |
-| QA investigators | **Sonnet** | `Teammate` tool via Agent Teams (spawned by qa-ruthless) |
+| QA investigators | **Sonnet** | `TeamCreate` + `Agent(implementer, team_name)` via Agent Teams (spawned by qa-ruthless) |
 | Code reviewer | **Opus** | `claude --agent reviewer` (evidence-based review, architecture judgment) |
 | Security auditor | **Opus** | `claude --agent security-auditor` (auth flows, tenant isolation, secrets, CVEs) |
 | Plan review | **Sonnet** | `plan-review` skill (constitution compliance needs reasoning) |
@@ -55,5 +55,5 @@ Rationale: Opus knows what to look for based on the feature. Haiku summaries los
 Constraint: Opus reads only files directly related to the feature scope.
 
 ## Implementer model
-Implementation teammates use **Sonnet** via the `Teammate` tool.
+Implementation teammates use **Sonnet** via Agent Teams (`TeamCreate` + `Agent(subagent_type: "implementer", team_name: ...)`).
 One teammate per repo. The teammate handles all commit units for its repo sequentially.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "cc-workspace",
-  "version": "5.2.1",
+  "version": "5.2.3",
   "description": "Claude Code multi-workspace orchestrator \u2014 skills, hooks, agents, and templates for multi-service projects",
   "bin": {
     "cc-workspace": "./bin/cli.js"