Executes complete autonomous development workflow from planning to knowledge capture. Use when user says "build [feature]", "implement [functionality]", "fix [bug]", "debug [issue]", "investigate [problem]", "optimize [component]", or asks technical questions requiring systematic solution. Handles feature development, debugging, investigation, and optimization with full plan → work → review → capture cycle.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Une commande directe contourne le prompt de vérification. Examinez la source avant de l'exécuter.
Executes complete autonomous development workflow from planning to knowledge capture. Use when user says "build [feature]", "implement [functionality]", "fix [bug]", "debug [issue]", "investigate [problem]", "optimize [component]", or asks technical questions requiring systematic solution. Handles feature development, debugging, investigation, and optimization with full plan → work → review → capture cycle.
Requirements clear? (yes/no)
- If yes: Auto-proceed
- If no: Ask clarifying questions
Example:
"Add user authentication" is AMBIGUOUS
- Which auth method? (OAuth, JWT, session)
- Registration included?
- Password requirements?
→ STOP and ask
Decision gate (Interactive):
Plan created: 8 tasks across 3 phases
[Show task breakdown]
Approve plan? (yes/no/modify)
Output:
Plan saved to .mycelium/plans/YYYY-MM-DD-{track_id}.md
Track ID stored in state.json
Plan registered in session_state.plans[] (auto-pauses any previously active plan)
Chains to Phase 3 (Implementation)
Error handling:
If planning fails: Report error, STOP
If user rejects plan: Iterate or exit
If requirements unclear: Ask questions, don't assume
Phase 3: Implementation
Invoked automatically by Phase 2 (via chain).
Pre-flight Checks
Verify baseline:
# Run existing tests
{test_command}
# Must all pass before new workif [ $? -ne 0 ]; thenecho"Baseline tests failing"echo"Fix existing tests before new work"
STOP
fi
Check uncommitted changes:
git status --porcelain
# If dirty working tree:# Autonomous: Stash changes# Interactive: Ask "Stash changes? (yes/no)"
Invoke TDD and Verification Skills
Load:
tdd (mandatory)
verification (mandatory)
Execute implementation:
For EACH task in plan (or parallel if dependencies allow):
Mark task in-progress:
[~] Task 1.1: Setup auth module
Search for patterns:
# Find similar implementations
grep -r "auth\|authentication" src/
# Check for learned patternscat .mycelium/solutions/patterns/critical-patterns.md
Apply TDD cycle (from tdd skill):
RED → GREEN → REFACTOR
1. Write failing test
2. Verify RED (test fails)
3. Write minimal implementation
4. Verify GREEN (test passes)
5. Refactor if needed
Verify with evidence:
# Run tests
{test_command}
# Show output# Verify exit code = 0# Check coverage
Commit:
git add {files}
git commit -m "Task 1.1: Setup auth module
- Add User model with password hashing
- Add JWT token generation
Tests: All pass (32/32)
Coverage: 85%
Co-Authored-By: Claude Sonnet 4.5 <noreply@anthropic.com>"
Workflow Complete!
Task: Add user authentication with JWT
Phase 0: Context Loading
✓ Project context loaded
✓ 15 solutions discovered
✓ 13 skills, 7 agents cached
Phase 1: Clarify Request
✓ Requirements clarified
✓ No external research needed
Phase 2: Planning & Assignment
✓ Created 8 tasks across 3 phases
✓ Capabilities assigned (agent/skills/model per task)
✓ TDD test strategy defined
Phase 3: Implementation
✓ 8/8 tasks completed
✓ All tests passing (45 tests)
✓ Coverage: 85% (target: 80%)
Phases 4.5 & 4.5B: Verification & Context Sync
✓ Tests verified with evidence
✓ Context managed (no bloat)
Phase 5: Review
✓ Spec compliance: PASS
✓ Quality score: 8.5/10
✓ 0 P1, 4 P2, 3 P3 issues
Phase 6: Finalization
✓ Commit: abc1234
✓ PR: #123 created
Phase 6E: Pattern Detection
✓ 2 patterns detected
✓ critical-patterns.md updated
Phase 6F: Store Knowledge
✓ 3 solutions stored
✓ 2 decisions documented
✓ Knowledge base updated
Next: Deploy or continue with next feature?
Next Steps
Suggest:
Merge to main (if on feature branch)
Create PR (if using pull requests)
Deploy (if ready)
Start next task: /mycelium-go [next-task]
State Management
Throughout execution, maintain state in state.json:
{"current_track":"auth_20260204","current_phase":"implementation","current_task":"1.3","mode":"autonomous","plans":[{"track_id":"auth_20260204","plan_file":"2026-02-04-auth_20260204.md","status":"in_progress","created":"2026-02-04T10:15:00Z","total_tasks":8,"completed_tasks":2}],"checkpoints":{"planning_complete":"2026-02-04T10:15:00Z","implementation_started":"2026-02-04T10:20:00Z","last_task_completed":"1.2","last_commit":"abc1234"},"blockers":[],"decisions_made":[{"decision":"Use JWT for authentication","rationale":"Stateless, scalable","timestamp":"2026-02-04T10:12:00Z"}]}
Save state:
After each phase completion (update plans[] entry with task progress)
After each task completion
Before asking user questions
On error/blocker
Resume support:
If interrupted, can resume from last checkpoint using /mycelium-continue. Use /mycelium-continue --full to force full workflow mode regardless of how the workflow was originally started.
Decision Matrix
When to STOP and Ask (Both Modes)
Scenario
Action
Example
Ambiguous requirements
ASK
"Add auth" → Which method?
High-risk change
ASK
Changing payment logic
P1 review issues
STOP
SQL injection found
Tests fail 3+ times
STOP
Cannot fix automatically
Missing credentials
ASK
Need API key
Architectural decision
ASK
Monolith vs microservices
When to AUTO-PROCEED (Autonomous Only)
Scenario
Action
Example
Clear requirements
GO
"Add JWT auth with bcrypt"
Pattern exists
GO
Found similar auth in codebase
Tests passing
GO
All green, proceed
P2/P3 issues only
GO
Note issues, continue
Standard task
GO
CRUD operations
Error Handling
Planning fails:
Planning failed: {reason}
Cannot proceed without plan.
Stopping.
Suggested: Clarify requirements and retry
Implementation blocked:
Implementation blocked: {reason}
Completed: 5/8 tasks
Blocked on: Task 1.6 (missing API key)
Saving state...
Resume with: /mycelium-continue
Review finds P1 issues:
Review FAILED: 2 critical (P1) issues
P1 Issues:
1. SQL injection vulnerability
2. Hardcoded credentials
These MUST be fixed before merge.
Stopping for fixes.
Knowledge capture fails:
Knowledge capture failed: {reason}
Work is complete but learnings not captured.
Options:
1. Retry capture: /mycelium-capture
2. Continue anyway (learnings lost)
Performance Optimization
Parallel execution:
Use worktrees for independent tasks
Dispatch multiple subagents
Merge results as they complete
Smart caching:
Load context once, reuse
Cache grep results
Reuse test runs if code unchanged
Progress streaming:
Update UI incrementally
Don't wait for phase completion
Show real-time progress
Skills Used
mycelium-plan: Requirements → tasks
tdd: RED → GREEN → REFACTOR enforcement
verification: Evidence-based validation
mycelium-review: Two-stage quality check
mycelium-capture: Knowledge extraction
Quick Examples
# Feature development
/mycelium-go "Add user authentication"# Debugging
/mycelium-go "Fix memory leak in session handler"# Investigation / Question
/mycelium-go "Why does the API return 500 on concurrent requests?"# Optimization
/mycelium-go "Optimize database queries" --interactive
# Detailed description
/mycelium-go "Add pagination to user list API with page size limits"
Resume with /mycelium-continue - If interrupted, /mycelium-continue resumes all remaining phases automatically (since /mycelium-go sets invocation_mode: "full")
--full flag - When resuming a single-phase skill (e.g., /mycelium-work), use /mycelium-continue --full to run all remaining phases instead of just finishing the current one
Result: Question answered with actionable solution and implementation completed.
Troubleshooting
Error: "Planning failed: Requirements unclear"
Cause: Ambiguous request with multiple valid interpretations
Solution: mycelium-go stops to ask clarifying questions. Answer them to proceed.
Error: "Tests failing after 3 attempts"
Cause: Implementation approach has fundamental issue
Solution: Check error logs in progress.md. Consider simplifying approach or using /recovery for systematic debugging.
Error: "Review FAILED: 2 critical (P1) issues"
Cause: Security vulnerabilities or critical bugs found
Solution: P1 issues must be fixed before merge. Review findings, implement fixes, re-run review.
Error: "No .mycelium directory found"
Cause: Project not initialized
Solution: Run /mycelium-setup first to bootstrap project structure.
Blocked: "Task 2.3 blocked: Waiting for API key"
Cause: External dependency missing
Solution: Provide required information, then run /mycelium-continue to resume.
