Execute ALL approved sprints sequentially and fully autonomously. Zero human intervention between sprints.
Execute ALL approved sprints sequentially and fully autonomously. Zero human intervention between sprints.
npx skills add https://github.com/de-braighter/workbench --skill all-sprints-runner이 명령을 Claude Code에 복사하여 붙여넣어 스킬을 설치하세요
| name | all-sprints-runner |
| description | Execute ALL approved sprints sequentially and fully autonomously. Zero human intervention between sprints. |
| tags | ["autonomous","sprint"] |
Execute all approved sprints sequentially, from first to last, fully autonomously.
Pre-condition: All sprint plans must be approved before running this skill. This skill removes the human gate between sprints — once started, it runs until every sprint is complete or a fatal error halts execution.
Read .claude/sdlc.json at the project root. Stop if missing.
All values below reference this config via sdlc.* notation.
NEVER STOP between sprints. After completing a sprint, IMMEDIATELY start the next one. Do NOT:
- Ask the user if you should continue
- Summarize progress and wait for confirmation
- Say "should I continue?" or "Sprint X is next in the queue"
- Cite context limits as a reason to pause
- Generate a "handoff state" for a new session
The ONLY reasons to stop are listed in Fatal errors below. Everything else = keep going.
After finishing Sprint N's retro, the very next action must be loading Sprint N+1's plan issue. If you run out of context, use
/context-minimizerto compress, or start a sub-agent for the next sprint. The user said "loop until done" — respect that.
Each sprint runs the full /sprint-runner lifecycle:
# Get all open approved sprint plan issues, sorted by sprint number
gh issue list --repo {sdlc.repo} \
--label "{sdlc.sprint.labels.plan}" --state open \
--json number,title --jq 'sort_by(.title | capture("Sprint (?<n>[0-9]+)").n | tonumber) | .[] | "\(.number) \(.title)"' \
--limit 50
Build the ordered sprint queue from the results.
git checkout main
git pull origin main
git status --porcelain # Must be empty
Run all verification commands from config (lint, test, build) to verify the codebase is green before starting.
If pre-flight fails: fix the issue before proceeding. The codebase must be clean at the start — zero errors AND zero warnings (if sdlc.lint.zeroWarnings is true).
for each SPRINT_PLAN in sprint_queue:
run_sprint(SPRINT_PLAN)
For each sprint, execute the full sprint-runner lifecycle (see /sprint-runner):
{sdlc.sprint.labels.backlog} to {sdlc.sprint.labels.current}For each story in execution order:
git checkout main && git pull origin main && git status --porcelain{sdlc.sprint.labels.blocked}, reorder if dependency not donegit checkout -b feature/{STORY_ID}-{short-description}sdlc.test.min thresholdssdlc.lint.commands, sdlc.test.commands, sdlc.build.commandssdlc.sprint.maxFixAttempts per step, then mark blocked and skipfeat({scope}): {STORY_ID} — {description}gh pr checks --watch)sdlc.sprint.reviewWaitSeconds, then check BOTH /pulls/{PR}/reviews AND /pulls/{PR}/commentssdlc.sprint.maxReviewRounds timessdlc.sprint.maxFixAttempts rounds)sdlc.lint.zeroWarnings)sdlc.test.min thresholdssdlc.i18n addedgh pr merge --squash --delete-branchgit checkout main && git pull origin mainAfter all stories in the sprint are done or skipped:
{sdlc.sprint.labels.techDebt} issues{sdlc.sprint.labels.backlog}, completed staysDO NOT summarize, ask the user, or generate a "handoff". Jump straight to the next sprint.
gh issue comment {NEXT_SPRINT_PLAN} --repo {sdlc.repo} \
--body "Auto-transitioning from Sprint {N} (velocity: {Y}pts). Beginning Sprint {N+1}."
Then without any pause or user message, start executing the next sprint.
After the last sprint is done:
gh issue create --repo {sdlc.repo} \
--title "All Sprints Complete — Full Product Build Summary" \
--label "type:refinement" \
--body "$(cat <<'EOF'
## Full Build Complete
All {total_sprints} sprints executed autonomously.
### Overall Metrics
- Total stories: {total_stories}
- Total points: {total_points}
- Stories completed: {completed}
- Stories skipped/blocked: {skipped}
- Tech debt items created: {debt_created}
- Tech debt items resolved: {debt_resolved}
- Average velocity: {avg_velocity} pts/sprint
### Sprint-by-Sprint Summary
| Sprint | Planned | Completed | Velocity | Debt Created | Debt Fixed |
|--------|---------|-----------|----------|--------------|------------|
| 1 | ... | ... | ... | ... | ... |
### Open Items
- Blocked stories requiring human intervention: ...
- Unresolved tech debt: ...
- Follow-up issues created: ...
EOF
)"
| Error | Recovery |
|---|---|
| Lint failure | Fix and re-run (up to sdlc.sprint.maxFixAttempts) |
| Test failure | Fix test or implementation, re-run |
| Build failure | Fix build error, re-run |
| Merge conflict | Pull main, rebase branch, resolve, push |
| Copilot requests changes | Apply fixes, push new commit |
| Git dirty working tree | Stash, return to main, pull, unstash if relevant |
| Network timeout on gh commands | Retry up to 3x with 10s delay |
| Error | Action |
|---|---|
| Max fix attempts failed | Mark story blocked, create investigation issue, skip |
| Max Copilot rounds exceeded | Escalate to human review, skip story |
| External dependency missing | Mark blocked, document blocker, skip |
| Story scope explosion | Implement what fits, create follow-up issue, close original |
| Error | Action |
|---|---|
| Main branch is broken (tests failing on main) | Stop all execution, create urgent issue |
| Git repository corrupted | Stop, alert |
| GitHub API consistently failing | Stop after 3 consecutive API failures |
On fatal error:
gh issue comment {CURRENT_SPRINT_PLAN} --repo {sdlc.repo} \
--body ":rotating_light: **FATAL: Sprint execution halted**
Error: {description}
Last completed story: {STORY_ID}
Stories remaining: {count}
Manual intervention required. Resume with \`/all-sprints-runner\` after fixing."
If execution is interrupted (crash, timeout, manual stop), running /all-sprints-runner again will:
This means the runner is idempotent — safe to restart at any point.
/all-sprints-runner # Run all remaining open sprints
No arguments needed. It discovers the queue automatically from open sprint plan issues.
Use when standing up a brand-new substrate domain in the de-braighter cluster — scaffold a building, testing, registered pnpm-workspace domain (reusable spine lib + pack lib + NestJS api), optionally with the DB-persistence, inference-backbone, and Angular-UI tiers. Codifies the markets reference run.
Use when setting up the de-braighter cluster on a new machine or fresh clone — the sibling layer/domain repos under layers/ and domains/ are missing, or their dependencies, the shared dev-Docker stack, or the domain databases aren't provisioned yet.
Scaffold a new ADR (enterprise or product) with correct numbering, template, and index update.
Single entry point that routes Angular and Nx architecture requests to the right workflow.
Optimize Angular and Nx code for fast localized updates with minimal ripple changes.
Enforce optimal Angular component decomposition in Nx workspaces.