| name | pro-workflow |
| description | Core AI coding workflow. Orchestrates 8 patterns: Scout → Plan → Implement → Review, with smart commits, session rituals, and continuous learning capture. This is the master skill — invoke it at the start of any non-trivial task.
|
| triggers | ["start working on","implement this feature","let's build","help me with this task","any task touching more than 1 file","any task with unclear scope"] |
| references | ["CLAUDE.md",".claude/agents/scout.md",".claude/agents/planner.md",".claude/agents/reviewer.md",".claude/state/"] |
Skill: pro-workflow
The 8 Core Patterns
| # | Pattern | Skill |
|---|
| 1 | Scout before you implement | scout agent |
| 2 | Plan before you code | implementation-planner |
| 3 | Test first | test-first-executor |
| 4 | Review before merge | council-review |
| 5 | Commit with quality gates | smart-commit |
| 6 | Persist corrections | learn-rule |
| 7 | Surface past patterns | replay-learnings |
| 8 | Hand off cleanly | wrap-up + session-handoff |
Workflow: Research → Plan → Implement
┌─────────────┐ ┌─────────────┐ ┌─────────────────┐
│ RESEARCH │────▶│ PLAN │────▶│ IMPLEMENT │
│ │ │ │ │ │
│ Scout scores│ │ Planner │ │ Phased execution│
│ readiness. │ │ breaks down │ │ Reviewer gates │
│ GO / HOLD │ │ steps. │ │ each checkpoint │
└─────────────┘ └─────────────┘ └─────────────────┘
│ │ │
Validation Validation Validation
gate: must gate: must gate: must
pass before pass before pass before
proceeding proceeding merge
Validation gate between each phase — must pass before proceeding.
Instructions
Phase 1 — Research (Scout)
- Invoke the Scout agent with the task description.
- If Scout returns HOLD: resolve each gap, then re-score.
- If Scout returns GO: proceed to Phase 2.
Scout checklist:
- [ ] Score ≥ 70 (GO verdict)
- [ ] Context mode selected (DEV / REVIEW / RESEARCH)
- [ ] No unresolved gaps
Phase 2 — Plan
- Run the
implementation-planner skill (or /plan command).
- Review the plan for risky modules — invoke
risky-module-review if flagged.
- Check
replay-learnings for relevant past patterns before accepting the plan.
Plan checklist:
- [ ] Plan produced with ordered steps
- [ ] Risky modules flagged and reviewed
- [ ] Past learnings consulted
- [ ] Max 10 steps (split if more)
Phase 3 — Implement
- Execute plan steps one at a time.
- After each file change, Reviewer agent validates before writing.
- Run
pytest -x after each logical group of changes.
- Run
deslop skill before final commit to remove AI code slop.
- Run
smart-commit to gate and commit.
Implement checklist:
- [ ] Reviewer passed each file change
- [ ] pytest -x green after each step
- [ ] docs/changelog.md updated
- [ ] deslop run before commit
- [ ] smart-commit used for final commit
Phase 4 — Wrap Up
- Run
council-review if changes are > 50 lines or touch risky modules.
- Run
wrap-up skill to close the session.
- Run
learn-rule to persist any corrections from this session.
- Run
session-handoff to write resume docs for the next session.
Model Selection Guide
Use the right Claude model for the task type:
| Task | Model |
|---|
| Quick fixes, typos, config changes | Haiku 4.5 |
| New features, standard implementations | Sonnet 4.6 (adaptive thinking) |
| Refactors, structural changes | Opus 4.6 (adaptive thinking) |
| Architecture decisions, ADR writing | Opus 4.6 (1M context) |
| Hard bugs, deep debugging | Opus 4.6 (1M context) |
Acceptance Checks
