Menu
Creates and manages dynamic teams of domain agents. Triggers: create team, agent team, team status, cleanup team.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
SSH server management — connect, configure, deploy, administer Linux servers with safety gates.
Manager mode. on installs+arms a HARD delegation wall into THIS project (PreToolUse denies Write/Edit/Bash in main session, subagents free); off disarms; uninstall removes it. Codeword ++m auto-injects a delegate-everything prompt (plan-aware: adds the plan supplement in plan mode); ++a auto-injects an architecture-first directive (mode-agnostic); ++rr/++r auto-inject review discipline (anti-regression / two-phase double-check), all independent of this skill. level strict|balanced, status, edit, reset. Triggers: manager, менеджер, hard mode, хард режим, delegate.
Installs or removes the think-short terse-mode hooks. Triggers: think-short, be terse, terse mode, инжект терс-режим.
Creates and improves Claude Code subagents. Triggers: create agent, improve agent, scaffold agent, fix agent.
Extracts etalon classes, patterns, architecture into convention docs. Triggers: extract conventions, etalon classes.
Orchestrates e2e testing: BDD scenarios, Playwright autotests. Triggers: e2e tests, BDD scenarios, write autotest.
| name | brewcode:teams |
| description | Creates and manages dynamic teams of domain agents. Triggers: create team, agent team, team status, cleanup team. |
| user-invocable | true |
| disable-model-invocation | true |
| argument-hint | [create [name] [prompt]|update [name]|status [name]|cleanup [name]] |
| allowed-tools | ["Read","Write","Edit","Glob","Grep","Bash","Task","AskUserQuestion","Skill"] |
| model | opus |
Manage dynamic teams of domain-specific agents with tracking framework.
Arguments: $ARGUMENTS
EXECUTE using Bash tool:
bash "${CLAUDE_SKILL_DIR}/scripts/detect-mode.sh" "$ARGUMENTS" && echo "OK" || echo "FAILED"
Output: MODE:, TEAM_NAME:, PROMPT: (optional). Store all three.
Output: Mode: {MODE}, Team: {TEAM_NAME}
Load environment:
| Action | Command / Path |
|---|---|
| Read agent template | ${CLAUDE_SKILL_DIR}/references/agent-template.md |
| Read framework templates | ${CLAUDE_SKILL_DIR}/references/framework-files.md |
| Check team dir | .claude/teams/{TEAM_NAME}/ -- exists? |
| Check existing agents | .claude/agents/ -- list all |
| If team.md exists | Read, show current roster |
| If trace.jsonl exists | Show entry counts via trace-ops.sh read |
If team exists, verify:
bash "${CLAUDE_SKILL_DIR}/scripts/verify-team.sh" "TEAM_NAME_HERE" && echo "PASS" || echo "FAIL"
Formulate action plan for current mode.
ASK using AskUserQuestion: "Here's my plan: {plan}. Continue?" Options: "Yes, continue" | "No, I want changes" | "Cancel"
Spawn 3-5 Explore agents in ONE message via Task tool:
| # | Focus |
|---|---|
| 1 | Code structure: modules, packages, domains, architectural layers |
| 2 | Existing agents (.claude/agents/, brewcode/agents/, ~/.claude/agents/) + Claude Code infrastructure |
| 3 | Tech stack: build files, frameworks, dependencies, languages |
| 4 | CI/CD, testing, deploy, infrastructure |
| 5 (optional) | Domain boundaries: business logic, API, data layer, UI |
All via Task(subagent_type="Explore"). Consolidate into single analysis document.
Based on analysis + PROMPT (if provided), propose 3 variants via AskUserQuestion:
Minimal (5 agents):
| Agent | Domain | Mission |
Balanced (10-12 agents) -- Recommended:
| Agent | Domain | Mission |
Maximum (15-20 agents):
| Agent | Domain | Mission |
Options: "Minimal (5)" | "Balanced (recommended)" | "Maximum (15-20)" | "Custom -- I'll specify"
If "Custom" -- second AskUserQuestion for free input. Final confirmation of agent list before proceeding.
"Default model for domain agents: Opus (most reliable)."
| Model | Best for | Cost |
|---|---|---|
| opus | Complex domains, architecture, critical logic | High |
| sonnet | Standard domains, CRUD, testing, utilities | Medium |
| haiku | Simple utility agents, formatting, validation | Low |
Options: "Opus (recommended)" | "Sonnet" | "Haiku" | "Mixed -- I'll choose per agent"
If "Mixed" -- ask model per agent in C3. Store as DEFAULT_MODEL (default: opus).
${CLAUDE_SKILL_DIR}/references/agent-template.mdTask(subagent_type="brewcode:agent-creator") with: placement=.claude/agents/, model=DEFAULT_MODEL (or per-agent), context=template + mission + domain + project analysis + colleague list. Agent description <= 100 chars (optimal ~80), single line, role + 2-3 triggers, no <example> blocks.Skill(skill="brewtools:text-optimize", args="-l .claude/agents/{agent-name}.md")
Create team directory:
mkdir -p ".claude/teams/TEAM_NAME_HERE" && echo "OK" || echo "FAILED"
Write from ${CLAUDE_SKILL_DIR}/references/framework-files.md templates: team.md (fill with real agent data), touch trace.jsonl
Verify:
bash "${CLAUDE_SKILL_DIR}/scripts/verify-team.sh" "TEAM_NAME_HERE" && echo "PASS" || echo "FAIL"
STOP if FAIL -- fix missing files before continuing.
AskUserQuestion: final report + suggest /brewcode:teams status {TEAM_NAME}
Spawn 3 reviewer agents in ONE message via Task tool:
| # | Focus |
|---|---|
| 1 | Instruction quality: clarity, imperative form, completeness, word budget |
| 2 | Domain accuracy: correct scope, tool selection, model fit, description triggers |
| 3 | Architecture: consistency across agents, no domain overlaps, proper Task Acceptance Protocol |
Each reads ALL agent files in .claude/agents/ and outputs:
FILE: .claude/agents/{name}.md
SEVERITY: critical/important/minor
ISSUE: description
FIX: suggested fix
Quorum threshold: 2/3 agreement = confirmed. Match criteria: same file + same area (+/- 5 lines or same section) + same category (instruction/domain/architecture/trigger).
| Outcome | Action |
|---|---|
| 2/3+ confirm | Mark confirmed, keep severity from highest reporter |
| 1/3 only | Log as unconfirmed, skip |
| Minor severity (all reporters) | Log but skip fix |
Task(subagent_type="brewcode:reviewer", prompt="
Verify these findings against actual agent files. For each:
1. Read the agent file
2. Check if the issue actually exists
3. Mark: VERIFIED or FALSE_POSITIVE
{confirmed_findings}
")
Filter out false positives. Final list = verified critical + important issues.
For each verified critical/important issue:
Task(subagent_type="brewcode:agent-creator", prompt="
Fix this issue in {agent_file}:
ISSUE: {description}
FIX: {suggested_fix}
SEVERITY: {severity}
Read the file, apply the fix, validate.
")
Batch: up to 3 parallel per message. Minor issues skipped.
Task(subagent_type="brewcode:reviewer", prompt="
Re-verify these fixes. For each:
1. Read the fixed agent file
2. Check original issue is resolved
3. Check no regression introduced
Mark: FIXED or REGRESSION
{fixes_applied}
")
| Outcome | Action |
|---|---|
| All FIXED | Pipeline complete, proceed to Epilogue |
| REGRESSION found | Return to C8 for that file (max 2 cycles) |
| Still failing after 2 cycles | Log as unresolved, proceed to Epilogue |
To skip review pipeline: add
--skip-reviewto create arguments. To run review on existing team:/brewcode:teams update {TEAM_NAME} --review
No modifications. Read + report only.
.claude/teams/{TEAM_NAME}/team.mdbash "${CLAUDE_SKILL_DIR}/scripts/trace-ops.sh" read ".claude/teams/{TEAM_NAME}" && echo "OK" || echo "FAILED"
src (agent) and k (kind). Compute per-agent stats from k=track (took/refused/completed/failed counts), issues from k=issue, insights from k=insight.Output:
# Team Status: {TEAM_NAME}
## Summary
| Metric | Value |
|--------|-------|
| Agents | {N} |
| Tasks tracked | {N} |
| Success rate | {%} |
| Open issues | {N} (high: {N}, critical: {N}) |
| Insights | {N} |
| Last activity | {date} |
## Per Agent
| Agent | Tasks | Success | Refused | Issues | Insights | Health |
|-------|-------|---------|---------|--------|----------|--------|
## Recommendations
Health:
| Label | Criteria |
|---|---|
| Healthy | >70% success, active |
| Needs tuning | 30-70% success or many refusals |
| Underperforming/Inactive | <30% success or inactive |
Recommendations: underperformers -> suggest /brewcode:teams update; >200 trace rows -> suggest cleanup; 0 activity -> suggest review.
No AskUserQuestion -- purely informational.
CURSOR=$(bash "${CLAUDE_SKILL_DIR}/scripts/trace-ops.sh" cursor ".claude/teams/{TEAM_NAME}")
bash "${CLAUDE_SKILL_DIR}/scripts/trace-ops.sh" read ".claude/teams/{TEAM_NAME}" --since "$CURSOR" && echo "OK" || echo "FAILED"
If cursor empty: all entries returned. If team not found -> STOP. If cursor exists and <10 post-cursor entries: expand to last 30 days.
Filter post-cursor trace: k=track for task stats, k=issue for problems, k=insight for patterns.
| Status | Criteria | Action |
|---|---|---|
| Healthy | >70% success, active | No changes |
| Needs tuning | 30-70% success or many refusals | Update instructions |
| Underperforming | <30% success | AskUser: update or delete+create new |
| Inactive | 0 records | AskUser: delete or keep |
ASK using AskUserQuestion with analysis table and proposed actions (Update/Delete/No changes per agent). Options: "Apply all" | "Let me choose" | "Show detailed analysis"
If "Let me choose" -> AskUserQuestion per agent. If "Show detailed" -> output full stats, then re-ask.
| Agent Status | Action |
|---|---|
| Needs tuning | Task(subagent_type="brewcode:agent-creator") update mode with tracking/issues/insights data |
| Underperforming (update) | Same as tuning |
| Underperforming (replace) | Delete agent file + create new via agent-creator |
| Inactive (delete) | Remove .claude/agents/{name}.md + update team.md status to removed |
Immutable traits (Name, Base Role) -> delete + create new. Mutable traits (Character, Instructions) -> update during tuning.
Update team.md with current state and Last update date.
Set cursor:
bash "${CLAUDE_SKILL_DIR}/scripts/trace-ops.sh" cursor ".claude/teams/{TEAM_NAME}" set "$(date -u +%Y-%m-%dT%H:%M:%SZ)" && echo "✅" || echo "❌ FAILED"
Read ${CLAUDE_SKILL_DIR}/references/cleanup-flow.md and execute step by step:
Archive: entries appended to .claude/teams/{TEAM_NAME}/trace-archive.jsonl. Cursor reset after cleanup.
Only for modes that change team composition (CREATE, UPDATE with removals, CLEANUP with agent removal):
ASK using AskUserQuestion: "Update team info in CLAUDE.md?" Options: "Yes, in project CLAUDE.md" | "Yes, in .claude/CLAUDE.local.md" | "No, skip"
Format to write:
## Teams
Team: {TEAM_NAME} | Agents: {N} | Status: active
| Agent | Domain | Mission |
|-------|--------|---------|
Protocol: agents self-select tasks, trace in `.claude/teams/{TEAM_NAME}/trace.jsonl`.
Manage: `/brewcode:teams [status|update|cleanup]`
Always run STATUS mode logic after all changes: read team.md + trace.jsonl, compute stats, output Team Status table.
# teams [{MODE}]
## Detection
| Field | Value |
|-------|-------|
| Arguments | `{raw args}` |
| Mode | `{MODE}` |
| Team | `{TEAM_NAME}` |
| Prompt | `{PROMPT or none}` |
## Results
{Mode-specific output}
## Next Steps
- {recommendations}
| Condition | Action |
|---|---|
| Team not found (STATUS/UPDATE/CLEANUP) | "Team '{TEAM_NAME}' not found. Run /brewcode:teams create {TEAM_NAME}." STOP |
| Team already exists (CREATE) | Show roster, AskUserQuestion: "Team exists. Update instead?" |
| verify-team.sh FAIL | Show missing items, attempt fix, re-verify |
| No agents created (C3 failure) | Retry failed agents once, then report |
| 0 trace entries (UPDATE) | Classify all agents as Inactive |