Autonomous Development
Description
Execute iterative development loops with automated quality gates, issue tracking integration, and intelligent remediation. Achieve 95/100 quality score before creating pull requests.
When to Use
- Large features broken into multiple issues
- Repetitive tasks that benefit from automation
- When consistent quality enforcement is needed
- Overnight/background development runs
Prerequisites
- GitHub or GitLab MCP configured
- Issues/epic created via
/plan --issues
.claude/autonomous.yml configured (or use defaults)
Quality Gates
Required Gates (Blockers)
| Gate |
Weight |
Threshold |
Remediation |
/test |
25 |
80% coverage, 0 failures |
Auto-fix test failures |
/security |
25 |
0 critical/high vulns |
Auto-fix vulnerabilities |
/build |
15 |
Build succeeds |
/build-fix |
Quality Gates
| Gate |
Weight |
Threshold |
Remediation |
/review |
20 |
< 3 smells |
/cleanup |
/mentor |
10 |
No critical concerns |
Create issues |
/ux |
5 |
A11y + responsive |
Create issues |
Scoring
Target: 95/100
Score = test + security + build + review + mentor + ux + bonus
Bonus:
+2 for coverage > 90%
+2 for zero code smells
+1 for zero security issues
Workflow
Phase 1: Planning
# Create epic with child issues
/plan "User authentication system" --issues
# Result:
# Epic #100: "User Authentication System"
# ├── Issue #101: "Create User model"
# ├── Issue #102: "Login endpoint"
# ├── Issue #103: "Session management"
# └── Issue #104: "Password reset"
Phase 2: Autonomous Execution
# Start autonomous development
/autonomous --epic 100 --target 95
# Or for a single issue
/autonomous --issue 101
Phase 3: Monitoring
Watch progress via:
- Issue comments (real-time updates)
- GitHub/GitLab notifications
/gate-status command
Phase 4: Review & Merge
- PRs created automatically when score >= 95
- Human reviews and approves
- On merge, loop continues to next issue
Configuration
Basic Config (.claude/autonomous.yml)
target_score: 95
max_iterations: 15
gates:
test:
weight: 25
required: true
thresholds:
min_coverage: 80
max_failures: 0
security:
weight: 25
required: true
thresholds:
critical_vulns: 0
high_vulns: 0
review:
weight: 20
auto_fix: true
Presets
| Preset |
Target |
Use Case |
default |
95 |
Standard development |
prototype |
80 |
Quick prototypes, MVPs |
production |
98 |
Production-critical code |
frontend |
95 |
UI components (UX required) |
/autonomous --epic 100 --preset production
Issue Integration
Issue Lifecycle
OPEN → IN PROGRESS → PR CREATED → IN REVIEW → MERGED → CLOSED
↓
(if blocked)
↓
BLOCKED → (human intervention) → IN PROGRESS
Labels Applied
| Label |
Meaning |
in-progress |
Autonomous loop working on issue |
blocked |
Loop paused, human needed |
ready-for-review |
PR created, awaiting approval |
auto-generated |
Issue created by autonomous run |
tech-debt |
Deferred improvement |
Sub-Issue Creation
Unresolved items become linked issues:
Issue #101: "Create User model" [CLOSED]
├── PR #110 [MERGED]
└── Sub-issues created:
├── #105: "[Security] Add rate limiting" (medium vuln)
└── #106: "[Tech Debt] Consider caching" (mentor recommendation)
Human Intervention
Via Issue Comments
| Command |
Action |
@claude [instruction] |
Execute instruction |
@pause |
Pause loop |
@resume |
Resume loop |
@skip-gate [gate] [reason] |
Skip a gate |
@target [score] |
Change target score |
Example
<!-- User comments -->
@claude Use bcrypt instead of argon2 for password hashing
<!-- Bot responds -->
@user Understood. Switching to bcrypt:
- Updated User model
- Modified password service
- Updated tests
Re-running quality gates...
Exit Conditions
Success (Create PR)
- Score >= 95 (or target)
- All required gates pass
Success with Review
- Score 85-94
- 10+ iterations attempted
- All required gates pass
- PR labeled
needs-review
Blocked (Pause)
- Required gate fails after max retries
- Score < 85 after max iterations
- Human intervention requested
Best Practices
Do
- Break features into small, focused issues
- Set realistic acceptance criteria
- Review autonomous PRs carefully
- Address tech-debt issues promptly
Don't
- Create overly complex single issues
- Ignore blocked notifications
- Auto-merge PRs (always review)
- Skip security gates without justification
Troubleshooting
| Problem |
Solution |
| Loop stuck on same score |
Check for flaky tests, review gate output |
| Too many sub-issues created |
Adjust thresholds in config |
| PR has conflicts |
Rebase branch, resume loop |
| Rate limited |
Increase iteration_delay |
Metrics
Track autonomous development effectiveness:
| Metric |
Target |
| Average score |
> 90 |
| Iterations per issue |
< 5 |
| Auto-fix success rate |
> 80% |
| Human interventions |
< 10% of issues |
Related
/plan - Create issues from feature request/orchestrate - Coordinate multiple autonomous runs/review - Code review gate/security - Security audit gate/mentor - Architecture review gate
liauw-media