skills/team-implement-plan/SKILL.md
Execute implementation plans with a small review team (Implementer + Reviewer + optional Integrator). The Implementer writes code directly while the Reviewer provides adversarial quality checks per phase. Use for moderate-to-complex plans where real-time adversarial review catches issues that automated checks miss. Triggers on "team implement", "team implement plan", or when the user explicitly wants team-based implementation.
npx skillsauth add mhylle/claude-skills-collection team-implement-planInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
This skill executes implementation plans using a small team of 2-3 members. Unlike the solo implement-plan which uses an orchestrator/subagent pattern, this skill has teammates that implement code directly and review each other's work adversarially.
Team composition:
When to use this vs implement-plan:
implement-plan for simple plans (1-3 phases, clear requirements, ~30-40K tokens/phase)team-implement-plan for quality-sensitive plans where adversarial review matters (~60-80K tokens/phase)team-implement-plan-full for large plans with parallel phases (~100-150K tokens/wave)Reference: See references/team-lifecycle.md for team lifecycle and references/quality-pipeline-distribution.md for pipeline distribution.
When invoked with a plan path:
"I'll set up a small implementation team. An Implementer will write code while a Reviewer independently checks quality. Let me read the plan and set up the team."
Step 1a: Read and validate the plan
Read($0) # Plan path from argument
Validate the plan has:
If the plan is missing required sections, inform the user and stop.
Step 1b: Check existing progress
TaskList # Check for existing tasks from this plan
If tasks exist with some completed, resume from the first incomplete phase. If no tasks exist, the plan was created without task bootstrapping — create tasks now.
Step 1c: Determine team size
| Plan Size | Team | |-----------|------| | 1-3 phases | Implementer + Reviewer (2 members) | | 4+ phases | Implementer + Reviewer + Integrator (3 members) |
Step 1d: Create the team
TeamCreate(team_name="impl-{plan-slug}")
Step 1e: Spawn teammates
Task(subagent_type="general-purpose",
team_name="impl-{plan-slug}",
name="implementer",
prompt="You are the Implementer on an implementation team.
PLAN: {full plan content}
CURRENT PHASE: {phase N details — objective, tasks, exit conditions}
YOUR ROLE: Implement this phase directly using Write, Edit, and Bash tools.
PROTOCOL:
1. Read all files mentioned in the plan's context section
2. Implement the phase tasks IN ORDER (tests first, then implementation)
3. After implementing, run all verification commands from exit conditions:
- Build verification: run build, lint, typecheck commands
- Runtime verification: start the app, check for errors
- Functional verification: run tests
4. Fix any failures — iterate until all exit conditions pass
5. When ALL exit conditions pass, message 'reviewer':
'Phase N implementation complete. Files changed: [list]. All exit conditions passing.'
RULES:
- Write code directly — you have full access to Write/Edit/Bash tools
- Follow existing codebase patterns (the plan documents them)
- Write tests BEFORE implementation code
- Do not proceed to the next phase — wait for Reviewer approval
- If you hit a blocker you cannot resolve, message the team lead
CURRENT TASK: Implement Phase {N} as described in the plan.")
Task(subagent_type="general-purpose",
team_name="impl-{plan-slug}",
name="reviewer",
prompt="You are the Reviewer on an implementation team.
PLAN: {full plan content}
YOUR ROLE: Independently verify each phase's implementation quality. You are the quality gate.
PROTOCOL:
1. Wait for the Implementer to message you that a phase is ready for review
2. When you receive a review request:
a. Read ALL changed files completely
b. Run integration tests (curl endpoints, check UI, verify behavior)
c. Apply code review checklist:
- Does the code follow existing patterns in the codebase?
- Are there any security issues? (injection, XSS, exposed secrets)
- Is error handling adequate?
- Are tests meaningful (not just trivial assertions)?
- Does the code match the plan's design decision?
d. Check ADR compliance: Read docs/decisions/INDEX.md, verify against relevant ADRs
e. Re-run all exit condition commands independently to confirm they pass
3. DECISION:
- If ALL checks pass: Message team lead with 'PASS: Phase N review complete. Quality assessment: [summary]'
- If ANY issues found: Message 'implementer' with 'NEEDS_CHANGES: [specific list of issues with file:line references]'
Then wait for implementer to fix and re-request review
RULES:
- Be thorough but pragmatic — flag real issues, not style preferences
- Verify independently — don't trust the Implementer's claim that tests pass
- Include file:line references for every issue
- Do NOT modify code yourself — only the Implementer writes code
- If you and the Implementer disagree on whether something is an issue, message the team lead")
Task(subagent_type="general-purpose",
team_name="impl-{plan-slug}",
name="integrator",
prompt="You are the Integrator on an implementation team.
PLAN: {full plan content}
YOUR ROLE: Monitor cross-phase consistency and handle plan synchronization.
PROTOCOL:
1. After each phase passes review, verify:
- No regressions from previous phases (run full test suite, not just phase tests)
- Integration between completed phases works correctly
- Shared resources (database schemas, API contracts, types) are consistent
- Module registrations, route files, index exports are updated
2. Update plan sync:
- Mark completed tasks via TaskUpdate
- Verify work items from the plan are actually done
- Flag any plan tasks that were skipped or partially completed
3. Report to team lead:
- 'INTEGRATION_PASS: Phase N integrates cleanly with previous phases'
- 'INTEGRATION_ISSUE: [description of cross-phase problem]'
RULES:
- Run the FULL test suite after each phase, not just new tests
- Check for import errors, missing exports, broken references
- If you find a regression, message 'implementer' with details")
For each phase in the plan:
Step 2a: Assign phase to Implementer
TaskUpdate(taskId="{phase-task-id}", status="in_progress")
SendMessage(type="message", recipient="implementer",
content="Begin Phase {N}: {phase name}.
Objective: {objective}.
Tasks: {task list}.
Exit conditions: {conditions}.
Files in scope: {files from plan}.")
Step 2b: Wait for implementation
Implementer works through the phase tasks. Monitor via TaskList. If Implementer goes idle without messaging Reviewer, prompt them.
Step 2c: Wait for review
Reviewer receives Implementer's "ready for review" message. Monitor the review.
Step 2d: Handle fix loop
If Reviewer sends NEEDS_CHANGES:
If fix loop exceeds 3 iterations, Lead intervenes:
Step 2e: Phase completion
When Reviewer sends PASS:
Step 2f: User confirmation
Present to user:
Phase {N} complete.
Implementation: {Implementer's summary}
Review: {Reviewer's assessment}
Integration: {Integrator's check, if applicable}
Files changed: {list}
Exit conditions: All passing
Continue to Phase {N+1}? (or /clear and resume later)
Wait for user confirmation before proceeding to next phase.
Step 2g: Re-assign for next phase
Update Implementer's context for the next phase:
SendMessage(type="message", recipient="implementer",
content="Phase {N} approved. Begin Phase {N+1}: {details}...")
After all phases complete:
Generate overall completion report:
Shut down teammates:
SendMessage(type="shutdown_request", recipient="implementer", ...)
SendMessage(type="shutdown_request", recipient="reviewer", ...)
SendMessage(type="shutdown_request", recipient="integrator", ...) # if exists
TeamDelete
Present final summary to user
If the session ends mid-implementation:
/team-implement-plan [plan-path]What persists: Task status (completed/pending), committed code, plan file What's lost: Teammate conversation context, in-progress uncommitted changes
Mitigation: Encourage committing after each phase passes review.
| Pipeline Step | Owner | Method | |---|---|---| | 1. Implementation | Implementer | Direct Write/Edit/Bash | | 2. Verification-loop | Implementer | Bash: build, lint, test commands | | 3. Integration testing | Reviewer | Bash: curl, test commands, behavioral checks | | 4. Code review | Reviewer | Read files, apply review checklist | | 5. ADR compliance | Reviewer | Read ADRs, check compliance | | 6. Plan sync | Integrator or Lead | TaskUpdate, verify work items | | 7. Prompt archival | Lead | Move prompt files if applicable | | 8. Completion report | Lead | Synthesize teammate reports |
Before completing each phase, verify:
Before completing the plan, verify:
tools
--- name: tt-workflow-build description: Tasktracker-native trigger for a PARALLEL build via the Claude Code Workflow tool. Thin by design — it does two things, then drives to done: (1) ensure a tasktracker project exists (use the existing one, or create one), then (2) start a dynamic `Workflow` that builds it, tracking the work in tasktracker and using the build + verify skills. It does NOT analyze parallelism up front, ask the user to choose a mode, hand back, or fall back to a sequential skil
tools
--- name: grumpy-reviewer description: A single grumpy, nitpicky structural code reviewer that runs as an isolated subagent and treats the code as third-party work submitted by a junior programmer for validation. It cares about exactly one thing — maintainability — judged through separation of concerns, service-oriented design, helper-method extraction, small files, and the rule of 7 (as any grouping nears 7 members, it pushes for sub-groupings). It is deliberately kept OUT of the implementation
development
--- name: tt-workflow-run description: Tasktracker-native autonomous build-loop orchestrator. Drives a first-class `workflow_run` end-to-end — create the run (Gate 1 lifecycle completeness + Gate 2 zero-defects-in), then loop while `getNextReadyTask(projectId)` returns a slice — `setActiveTask` → record a pre-slice `scanArchitectureDrift` baseline → delegate the slice to `/tt-implement-phase` (which does the code work, registers the architecture delta in-slice, and auto-logs defects/learnings/fr
tools
Tasktracker-native project-wide parallel audit using the Claude Code Workflow tool (dynamic workflows). Partitions a repo / backlog / architecture and fans out read-only agents (one per partition) that return schema-checked findings, aggregates them into a deduplicated, ranked risk register, and OPTIONALLY writes fixes back as tasks under a Bug Fix phase — with all tasktracker writes done by the PARENT, never the parallel agents (single global active-task pointer). Journaled and resumable, so a rate-limit or crash mid-audit resumes without re-running completed partitions. Use for large, embarrassingly-parallel, read/analyze-heavy jobs where each unit is self-contained and the output aggregates — audit every file/component for risk, find all architecture drift (scanArchitectureDrift) or duplicate tasks (detectDuplicates/auditDuplicates), per-file tech-debt sweep, test-coverage or security-surface scan across a whole project. Triggers on "/tt-workflow-audit", "audit the whole repo", "parallel audit", "scan every file/component", "find all drift/duplicates", "tech-debt sweep (tasktracker)", or any whole-project analyze-at-scale request inside a session with a tasktracker project. Prefer this over /codebase-audit or /code-quality-audit when the project is tracked in tasktracker AND you want the findings written back as tasks; prefer it over team-* modes when the units don't need to negotiate live (they just report).