13 KiB
Workflow Templates
Copy/paste templates for common multi-skill workflows. Each workflow uses the shared conventions pattern from the main skill.
Table of Contents
- Triage → Plan → Implement → Test → Review → Ship
- Spec Gate → Implement → Security Review → Merge
- PR Summary → Review Notes → Update PR
- Repo Bootstrap → Conventions → First Task
- Incident Triage → Evidence → Hypothesis → Fix → Postmortem
- Data Report → Visualize → Publish
- Council Review → Decision → Implementation
- Safe Refactor Loop
- Doc-Driven Development
- Release Workflow
Triage → Plan → Implement → Test → Review → Ship
The canonical development workflow. Use for feature work, bug fixes, and improvements.
Structure
.claude/skills/
triage/SKILL.md
plan/SKILL.md
implement/SKILL.md
test/SKILL.md
review/SKILL.md
ship/SKILL.md
artifacts/
triage.md
plan.md
test-report.md
review-notes.md
triage/SKILL.md
---
name: triage
description: Turn incoming task into problem statement + acceptance criteria.
context: fork
agent: Explore
allowed-tools: Read, Grep, Glob, Write
---
Triage $ARGUMENTS.
Write artifacts/triage.md:
- Problem statement
- Suspected scope (files/modules)
- Acceptance criteria
- Risks + unknowns
plan/SKILL.md
---
name: plan
description: Convert triage into implementation plan with checkpoints and rollback.
disable-model-invocation: true
---
Read artifacts/triage.md and constraints.md.
Write artifacts/plan.md:
- Approach (1-2 options)
- Chosen option + rationale
- Task breakdown
- Test plan
- Rollback plan
implement/SKILL.md
---
name: implement
description: Implement the planned changes following artifacts/plan.md.
disable-model-invocation: true
---
Follow artifacts/plan.md.
- Make minimal diffs
- Update context.md with decisions
- Prefer small commits
test/SKILL.md
---
name: test
description: Run test plan and summarize failures deterministically.
disable-model-invocation: true
allowed-tools: Read, Bash, Write
---
Run commands from artifacts/plan.md "Test plan".
Write artifacts/test-report.md:
- Commands run
- Output summary
- Failures and fixes
review/SKILL.md
---
name: review
description: Self-review like a strict PR reviewer. Propose follow-ups.
context: fork
agent: Plan
allowed-tools: Read, Grep, Glob, Write
---
Review diffs and artifacts.
Write artifacts/review-notes.md:
- Risks
- Edge cases
- Refactor opportunities
ship/SKILL.md
---
name: ship
description: Finalize and ship. Only when explicitly invoked.
disable-model-invocation: true
---
Checklist:
- tests green
- artifacts complete
- review notes addressed
Then perform ship steps appropriate to this repo.
State Flow
/triage → artifacts/triage.md
↓
/plan reads triage.md → artifacts/plan.md
↓
/implement reads plan.md → code changes + context.md
↓
/test reads plan.md → artifacts/test-report.md
↓
/review reads all → artifacts/review-notes.md
↓ (gates /ship)
/ship reads review-notes.md → commit/PR/deploy
Spec Gate → Implement → Security Review → Merge
Adds adversarial security review before merge. Use for security-sensitive features.
Structure
.claude/skills/
spec-gate/SKILL.md
implement/SKILL.md
adversarial-review/SKILL.md
merge/SKILL.md
artifacts/
spec.md
security-review.md
spec-gate/SKILL.md
---
name: spec-gate
description: Write spec and detect prompt-injection/scope ambiguity before coding.
context: fork
agent: Plan
allowed-tools: Read, Grep, Glob, Write
---
Write artifacts/spec.md:
- Goals / non-goals
- Constraints
- Acceptance criteria
- Threat model (where could data leak?)
adversarial-review/SKILL.md
---
name: adversarial-review
description: Adversarial review against prompt injection and unsafe tool use.
context: fork
agent: Plan
allowed-tools: Read, Grep, Glob, Write
---
Review diff and artifacts/spec.md.
Write artifacts/security-review.md:
- Suspicious instructions
- Risky tool calls
- Recommended restrictions (allowed-tools / hooks)
PR Summary → Review Notes → Update PR
Live PR workflow using gh CLI preprocessing.
pr-summary/SKILL.md
---
name: pr-summary
description: Summarize current PR using live gh CLI output.
context: fork
agent: Explore
allowed-tools: Read, Bash(gh:*), Write
---
## Pull Request Context
- **Diff**: !`gh pr diff`
- **Comments**: !`gh pr view --comments`
Summarize changes and risks in artifacts/pr-summary.md.
review-notes/SKILL.md
---
name: review-notes
description: Generate review notes from PR summary.
context: fork
agent: Plan
allowed-tools: Read, Write
---
Read artifacts/pr-summary.md.
Write artifacts/review-notes.md:
- Key changes
- Concerns
- Questions for author
update-pr/SKILL.md
---
name: update-pr
description: Update PR description with generated summary.
disable-model-invocation: true
allowed-tools: Read, Bash(gh:*)
---
Read artifacts/pr-summary.md.
Update PR body using gh pr edit.
Repo Bootstrap → Conventions → First Task
Onboarding workflow for new projects.
bootstrap-repo/SKILL.md
---
name: bootstrap-repo
description: Initialize .claude/ structure for new project.
disable-model-invocation: true
---
Create skeleton:
- .claude/skills/_shared/context.md
- .claude/skills/_shared/constraints.md
- artifacts/ directory
Populate constraints.md with project defaults.
conventions/SKILL.md
---
name: conventions
description: Fill in repo-specific conventions after bootstrap.
---
Read existing codebase patterns.
Update constraints.md with:
- Style conventions
- Testing requirements
- Security policies
Incident Triage → Evidence → Hypothesis → Fix → Postmortem
Incident response workflow with deterministic evidence gathering.
incident-triage/SKILL.md
---
name: incident-triage
description: Initial incident assessment and severity classification.
context: fork
agent: Explore
allowed-tools: Read, Grep, Glob, Write
---
Assess:
- Symptoms
- Affected systems
- Severity level
- Initial timeline
Write artifacts/incident-triage.md.
gather-evidence/SKILL.md
---
name: gather-evidence
description: Collect logs and metrics deterministically.
context: fork
agent: Explore
allowed-tools: Read, Bash(grep:*), Bash(tail:*), Write
---
## Current State
- **Recent logs**: !`tail -100 /var/log/app.log`
- **Error count**: !`grep -c ERROR /var/log/app.log`
Write artifacts/evidence.md with findings.
hypothesize/SKILL.md
---
name: hypothesize
description: Form and rank hypotheses from evidence.
context: fork
agent: Plan
allowed-tools: Read, Write
---
Read artifacts/evidence.md.
Write artifacts/hypothesis.md:
- Hypotheses ranked by likelihood
- Evidence supporting each
- Investigation steps to confirm/reject
postmortem/SKILL.md
---
name: postmortem
description: Generate postmortem from incident artifacts.
context: fork
agent: Plan
allowed-tools: Read, Write
---
Read all incident artifacts.
Write artifacts/postmortem.md:
- Timeline
- Root cause
- Contributing factors
- Action items
- Lessons learned
Data Report → Visualize → Publish
Reporting workflow with artifact generation.
data-report/SKILL.md
---
name: data-report
description: Gather and analyze data for report.
context: fork
agent: Explore
allowed-tools: Read, Bash(*), Write
---
Gather data per $ARGUMENTS.
Write artifacts/data-report.md with analysis.
visualize/SKILL.md
---
name: visualize
description: Generate visualizations from report data.
allowed-tools: Read, Bash(*), Write
---
Read artifacts/data-report.md.
Generate charts/diagrams in artifacts/visuals/.
publish-report/SKILL.md
---
name: publish-report
description: Compile final report for publishing.
disable-model-invocation: true
---
Combine artifacts/data-report.md and artifacts/visuals/.
Output final report to artifacts/final-report.md or HTML.
Council Review → Decision → Implementation
Multi-perspective review pattern. Forces diverse failure modes.
council-review/SKILL.md
---
name: council-review
description: Gather multiple perspectives on a decision.
context: fork
agent: Plan
allowed-tools: Read, Grep, Glob, Write
---
Review $ARGUMENTS from perspectives:
- Security reviewer
- Performance reviewer
- UX/Product reviewer
- Maintainability reviewer
Write artifacts/council-review.md with each perspective.
decision/SKILL.md
---
name: decision
description: Synthesize council review into decision.
---
Read artifacts/council-review.md.
Write artifacts/decision.md:
- Chosen approach
- Rationale
- Dissenting views acknowledged
- Mitigations for concerns
Safe Refactor Loop
Read-only exploration before any changes.
explore-safe/SKILL.md
---
name: explore-safe
description: Read-only codebase exploration.
context: fork
agent: Explore
allowed-tools: Read, Grep, Glob, Write
---
Explore $ARGUMENTS without making changes.
Write artifacts/exploration.md with findings.
refactor-plan/SKILL.md
---
name: refactor-plan
description: Plan refactoring from exploration findings.
context: fork
agent: Plan
allowed-tools: Read, Write
---
Read artifacts/exploration.md.
Write artifacts/refactor-plan.md:
- Changes needed
- Order of operations
- Test coverage requirements
- Rollback strategy
refactor-execute/SKILL.md
---
name: refactor-execute
description: Execute refactoring plan.
disable-model-invocation: true
---
Follow artifacts/refactor-plan.md exactly.
Run tests after each step.
Doc-Driven Development
Outline → Spec → Code → Docs Sync
outline/SKILL.md
---
name: outline
description: Create high-level outline before specifying.
context: fork
agent: Plan
allowed-tools: Read, Write
---
Write artifacts/outline.md with structure and goals.
spec/SKILL.md
---
name: spec
description: Detailed specification from outline.
---
Read artifacts/outline.md.
Write artifacts/spec.md with full specification.
docs-sync/SKILL.md
---
name: docs-sync
description: Sync documentation with implementation.
---
Compare code to artifacts/spec.md.
Update documentation to match implementation.
Release Workflow
Preflight → Build → Deploy (manual) → Verify → Announce
preflight/SKILL.md
---
name: preflight
description: Pre-release validation checklist.
allowed-tools: Read, Bash(*), Write
---
Run preflight checks:
- Tests pass
- Lint clean
- No security warnings
- Changelog updated
Write artifacts/preflight.md with results.
Block if any checks fail.
deploy/SKILL.md
---
name: deploy
description: Deploy to environment. Manual invocation only.
disable-model-invocation: true
allowed-tools: Read, Bash(*)
---
Read artifacts/preflight.md (must exist and pass).
Deploy to $ARGUMENTS environment.
verify/SKILL.md
---
name: verify
description: Post-deploy verification.
allowed-tools: Read, Bash(*), Write
---
Verify deployment health:
- Health endpoints responding
- Key flows working
- No error spikes
Write artifacts/verify.md with results.
announce/SKILL.md
---
name: announce
description: Announce release. Manual invocation only.
disable-model-invocation: true
---
Read artifacts/verify.md (must pass).
Generate release announcement.
Pattern Summary
| Workflow | Key Insight |
|---|---|
| Triage→Ship | Full development lifecycle with gates |
| Spec Gate | Adversarial security review before merge |
| PR Summary | Preprocessing with !gh for live context |
| Bootstrap | Onboarding pattern for new projects |
| Incident | Evidence-first debugging with postmortem |
| Data Report | Artifact generation with visualization |
| Council | Multi-perspective review forces diverse analysis |
| Safe Refactor | Read-only exploration before changes |
| Doc-Driven | Spec precedes code |
| Release | Manual gates for deploy/announce |
The "secret sauce" isn't the step names—it's the state handoff discipline via artifacts.