Multi-agent workflow orchestration patterns for coordinating test-engineer, code-reviewer, supervisor, and doc-writer agents. Use when managing complex coding tasks that require multiple specialized agents.
This skill is limited to using the following tools:
This skill provides patterns for orchestrating multiple specialized agents in a coordinated workflow.
There is NO iteration limit. Continue fixing and re-testing until:
The #1 cause of workflow failure is losing sight of the original requirement.
To prevent this:
| Agent | Role | When to Invoke |
|---|---|---|
| test-engineer | Write and run tests | After code implementation |
| code-reviewer | Review code quality and security | After tests pass |
| supervisor | Prevent context drift and regression | Every 3 iterations, when issues increase, before completion |
| doc-writer | Generate documentation | When docs needed (optional) |
When delegating to an agent, provide:
Use the **{agent-name}** agent to {task description}.
**Original Requirement (EXACT QUOTE):**
"{user's original words}"
**Context:**
- Previous step: {what was done}
- Files modified: {list of files}
- Current iteration: {N}
**Scope:**
- {specific files or modules}
**Requirements:**
- {deliverable 1}
- {deliverable 2}
**Constraints:**
- Follow CLAUDE.md standards
- {other constraints}
**Wait for agent completion before proceeding.**
+-------------------------------------------------------------------+
| INFINITE ITERATION LOOP |
| (Continue until PERFECT completion) |
+-------------------------------------------------------------------+
| |
| Implement --> test-engineer --> Tests pass? |
| ^ | |
| | +----------+----------+ |
| | Yes No |
| | | | |
| | v +--> Fix ---------+
| | code-reviewer |
| | | |
| | Approved? |
| | +-------+-------+ |
| | Yes No |
| | | | |
| | v +--> Fix --> test-engineer -----+
| | (Every 3 iterations) |
| | | |
| | v |
| | supervisor --> CONTINUE? |
| | | |
| | +-------+-------+-------+ |
| | Yes PAUSE RESET |
| | | | | |
| | v v v |
| | ALL MET? Ask User Revert and restart |
| | | |
| +----+ No |
| | |
| Yes |
| | |
| v |
| COMPLETE |
| |
| SAFEGUARD: supervisor PREVENTS CONTEXT DRIFT AND REGRESSION |
| |
+-------------------------------------------------------------------+
Before completing, verify ALL of the following:
[ ] User requirement fully implemented? (check against ORIGINAL REQUIREMENT)
[ ] ALL tests passing (100%)?
[ ] Code review APPROVED?
[ ] No security vulnerabilities?
[ ] No bugs or regressions?
[ ] Follows project standards (CLAUDE.md)?
[ ] Supervisor APPROVED?
If ANY checkbox is NOT checked, continue iterating. DO NOT STOP.
Invoke the supervisor agent when:
If CONTINUE:
If PAUSE:
If RESET:
Starting Droids workflow for: {requirement}
ORIGINAL REQUIREMENT (PRESERVED):
"{exact user words}"
Workflow steps:
1. [PENDING] Analyze and implement
2. [PENDING] Test with test-engineer
3. [PENDING] Review with code-reviewer
4. [PENDING] Verify with supervisor
5. [PENDING] Document (if needed)
Note: Will iterate until ALL criteria are met. No iteration limit.
Safeguard: supervisor will prevent context drift and regression.
Iteration {N} Complete
Test Results: {X/Y passed}
Review Status: {APPROVED/NEEDS CHANGES}
Issues Found: {count}
Trend: {IMPROVING/STABLE/DEGRADING}
Next Action: {what will be done}
Original Requirement (reminder):
"{exact user words}"
## Workflow Complete
### Original Requirement
"{exact user words}"
### Summary
- Implemented: {features}
- Total iterations: {count}
- Tests: ALL PASSED ({coverage}%)
- Review: APPROVED
- Supervisor: APPROVED
- Documentation: {status}
### Verification Checklist
[x] User requirement fully implemented
[x] All tests passing
[x] Code review approved
[x] No security vulnerabilities
[x] No bugs
[x] Follows project standards
[x] Supervisor approved
### Files Modified
- {file list}
Use the **test-engineer** agent to create and run comprehensive tests.
**Original Requirement:**
"{exact user words}"
**Context:**
- Implemented user authentication feature
- Files: src/auth/login.ts, src/auth/register.ts
- Current iteration: 1
**Requirements:**
- Unit tests for all functions
- Integration tests for API endpoints
- Coverage report
**Wait for completion. If tests fail, fix and re-invoke until ALL pass.**
Use the **code-reviewer** agent to assess code quality.
**Original Requirement:**
"{exact user words}"
**Context:**
- Feature: User authentication
- Tests: 15/15 passed (92% coverage)
- Current iteration: 1
**Requirements:**
- Security review (OWASP top 10)
- Code quality assessment
- Performance analysis
- CLAUDE.md compliance check
**Wait for completion. If issues found, fix ALL, re-test, and re-review.**
Use the **supervisor** agent to check requirement alignment and detect regression.
**Original Requirement (EXACT):**
"{exact user words}"
**Iteration History:**
- Iteration 1: 10/15 tests passed, 5 review issues
- Iteration 2: 13/15 tests passed, 3 review issues
- Iteration 3: 15/15 tests passed, 1 review issue
**Files Modified:**
- src/auth/login.ts
- src/auth/register.ts
- tests/auth.test.ts
**Current Status:**
- Tests: ALL PASSING
- Review: 1 minor suggestion remaining
**Request:**
- Verify we are still aligned with original requirement
- Check for any regression
- Recommend: CONTINUE / PAUSE / RESET
Use the **doc-writer** agent to generate documentation.
**Context:**
- Feature: User authentication (implemented, tested, reviewed, supervisor approved)
- Language: {user's language}
**Requirements:**
- API documentation for new endpoints
- JSDoc comments for public functions
- Update README if needed
**Respond in the same language as the user.**