| name | dev-cycle |
| description | Orchestrate the full GitHub-issues-driven development lifecycle. 7-phase pipeline from brainstorm through PR with state tracking and cross-conversation resume. Use when user says "dev cycle", "development workflow", "full development pipeline", or invokes /dev-cycle.
|
Dev Cycle Orchestrator
Orchestrate the full development lifecycle: brainstorm → plan → review → issues → implement → code review → PR.
Disambiguation: If the user only wants a PRD, route to /write-a-prd. If they only want a plan, route to /prd-to-plan. This skill is for the full end-to-end lifecycle.
The 7-Phase Pipeline
Every phase is mandatory. No phase can be skipped.
| # | Phase | Delegates To | Gate Condition |
|---|
| 1 | Brainstorm | write-a-prd | Issue URL recorded |
| 2 | Plan | prd-to-plan | Plan file exists at docs/plans/ |
| 3 | CEO Review | plan-ceo-review (recommend HOLD SCOPE) | Review complete, user approves |
| 4 | Issues | Orchestrator (plan slices → GitHub issues) | All issue URLs recorded |
| 5 | Implement | Orchestrator (tdd per issue, subagent-development) | All issues resolved, tests pass |
| 6 | Code Review | daa-code-review | Clean review |
| 7 | PR | commit + github-cli | PR URL recorded |
Re-entry Logic
On every invocation:
No arguments (/dev-cycle)
- Scan
docs/dev-cycle/ for *.state.md files with status: in_progress
- If one → ask: "Resume {feature} (
{branch})? Currently at {phase}."
- If multiple → list with branch names, ask which to resume
- If none → ask: "What feature are you working on?" → start Phase 1
With argument (/dev-cycle {slug})
- Look for
docs/dev-cycle/{slug}.state.md
- Found → resume at
current_phase
- Not found → create state file, start Phase 1
Slug Collision
When creating a new state file, check docs/dev-cycle/ for existing slugs. If the slug already exists (abandoned or completed), suffix with -2, -3, etc.
Context Loading on Resume
Before continuing, load ALL referenced artifacts:
- Brainstorm:
gh issue view the PRD issue
- Plan: Read plan file from disk
- CEO Review: Read the plan file (includes review revisions)
- Issues:
gh issue view each implementation issue
- Implement: Check git status on feature branch, review closed issues
Present summary: "Resuming {feature} at {phase}. Here's where we left off: ..."
Phase Execution
Phase 1: Brainstorm
Invoke write-a-prd. Trust the skill's internal flow (interview → PRD → GitHub issue). Record the issue URL in the state file.
Phase 2: Plan
Pass the PRD issue URL to prd-to-plan. Record the plan file path (at docs/plans/{feature}.md).
Phase 3: CEO Review
Pass plan file path to plan-ceo-review. Recommend HOLD SCOPE mode but let the skill's own mode selection (Step 0F) run. Record when review is complete and user approves.
Phase 4: Issues
Owned by the orchestrator. Read the plan's vertical slices. For each slice, create a GitHub issue using gh issue create that:
- References the PRD issue
- Includes acceptance criteria from the plan
- Is created in dependency order
Record each issue URL in the Issues table immediately after creation. See references/phase-transitions.md for partial-completion recovery.
Phase 5: Implement
Owned by the orchestrator. Create feature branch feat/{feature-slug} (see branch handling in references/phase-transitions.md).
Dispatch one subagent per GitHub issue following subagent-development methodology:
- Each subagent invokes the
tdd skill
- Code review runs between each subagent dispatch
- State file is updated after each subagent completes (not batched)
Log per-subagent events:
"Subagent started for issue #N: {title}"
"Subagent completed for issue #N: {pass/fail}"
"Code review after issue #N: {clean/blocking issues found}"
Phase 6: Code Review
Invoke daa-code-review against all changed files on the feature branch. If blocking issues found → fix, re-run. Loop until clean.
If architectural issues requiring plan rework → trigger backwards transition to Phase 2.
Phase 7: PR
Check for conflicts with default branch first. Invoke commit for conventional commit, then github-cli to open PR. Record PR URL, set status: completed. Then run the archival step (see Archival below).
State File
Each feature tracked at docs/dev-cycle/{feature-slug}.state.md. See references/state-file-schema.md for full format, field definitions, and transition rules.
Failure & Recovery
See references/phase-transitions.md for:
- Phase retry logic (blocked → retry on next invocation)
- Backwards transitions (implement/code_review → plan)
- Feature abandonment. Archived files are moved to
docs/archive/.
Branch Management
- Phases 1–4: Run on current branch (documentation only)
- Phase 5: Creates
feat/{feature-slug} branch
- Phase 7: PRs to the default branch (detected via
gh repo view --json defaultBranchRef)
- Resume at Phase 5+: Check out feature branch if not already on it
Archival
When a feature reaches a terminal state (completed or abandoned), archive its artifacts:
- Create archive directories:
mkdir -p docs/archive/dev-cycle docs/archive/plans
- Move the state file:
git mv docs/dev-cycle/{slug}.state.md docs/archive/dev-cycle/
- Move the plan file: read the plan path from the artifacts table, then
git mv {plan_path} docs/archive/plans/
- Commit the moves with message:
chore(dev-cycle): archive {slug}
Archival runs automatically:
- At the end of Phase 7 (after PR URL is recorded and status is set to
completed)
- On feature abandonment (after status is set to
abandoned)