Help us improve
Share bugs, ideas, or general feedback.
From autonomous-dev
Internal skill for overnight autonomous execution. Provides context and rules for unattended development.
npx claudepluginhub adiomas/claude-code-adiomas-plugins --plugin autonomous-devHow this skill is triggered — by the user, by Claude, or both
Slash command
/autonomous-dev:skills/overnight-modeThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
You are currently running in **OVERNIGHT MODE**. This skill provides essential context for unattended autonomous development.
Automates overnight software development using Git hooks to enforce TDD, ensuring tested code changes meet quality standards before commits.
Executes TDD-based autonomous development loops for tasks with checkpoint recovery, resume, status checks, iteration limits, and observability changelog.
Auto-loop execution workflow with quality gates. Use when starting any non-trivial implementation task. Provides automatic task decomposition, code implementation, testing (L1-L4), and iterative quality gates until completion. Invoke with /autoworker.
Share bugs, ideas, or general feedback.
You are currently running in OVERNIGHT MODE. This skill provides essential context for unattended autonomous development.
Read your current state from these files:
.claude/auto-overnight.local.mdactive: true/false # Is overnight mode active?
iteration: N # Current loop iteration
max_iterations: N # Stop after this many
max_hours: N # Maximum runtime
started_at: ISO8601 # When overnight started
deadline_at: ISO8601 # Hard stop time
current_phase: PHASE # INIT/PLAN/EXECUTE/VERIFY/COMPLETE
tasks_completed: N # How many tasks done
tasks_total: N # Total tasks in plan
last_checkpoint: ISO # Last checkpoint time
.claude/auto-progress.yamltasks:
task-1:
status: done/in_progress/pending/failed
branch: auto/task-1
iterations: N
.claude/auto-memory/context-summary.md - Full context for resumeovernight-checkpoint.md - Latest overnight checkpointnext-actions.md - What to do nextAskUserQuestion tool.claude/overnight-decisions.mdEven overnight, you MUST follow TDD:
RED → Write ONE failing test FIRST
GREEN → Implement MINIMAL code to pass
REFACTOR → Clean up while tests pass
MUTATE → Verify test quality (if tools available)
After every significant action:
# Update checkpoint timestamp
TIMESTAMP=$(date -u +%Y-%m-%dT%H:%M:%SZ)
sed -i '' "s/last_checkpoint: .*/last_checkpoint: \"$TIMESTAMP\"/" .claude/auto-overnight.local.md
Never claim "tests pass" without running them:
# WRONG
echo "Tests should pass"
# RIGHT
npm test 2>&1 | tee .claude/overnight-test-output.log
echo "Test output saved to .claude/overnight-test-output.log"
If context is filling up:
.claude/auto-memory/overnight-checkpoint.md/auto-continue.claude/overnight-issues.md and continue.claude/overnight-issues.mdThe stop hook will use prompt-based LLM evaluation to check:
All plan tasks completed?
.claude/auto-progress.yamldoneAll verification passing?
No critical issues?
.claude/overnight-issues.md has no unresolved CRITICAL itemsOnly when ALL three are true, overnight mode completes.
When complete, generate .claude/overnight-report-{timestamp}.md:
# Overnight Development Report
## Execution Summary
- **Started:** {started_at}
- **Completed:** {timestamp}
- **Duration:** {duration}
- **Iterations:** {iteration}
- **Tasks:** {completed}/{total}
## Changes Made
| File | Action | Lines |
|------|--------|-------|
| src/foo.ts | Created | +150 |
| src/bar.ts | Modified | +30/-10 |
## Verification Results
### Tests
\`\`\`
{actual npm test output}
\`\`\`
### Lint
\`\`\`
{actual npm run lint output}
\`\`\`
### Build
\`\`\`
{actual npm run build output}
\`\`\`
## Issues Encountered
{contents of overnight-issues.md or "None"}
## Git History
\`\`\`
{git log --oneline for overnight commits}
\`\`\`
## Recommendations
{any follow-up work or concerns}
If overnight mode ends with issues:
.claude/overnight-report-*.md.claude/overnight-issues.md/auto-continue to resume if needed