minhtran3124/agent-harness/writing-plans
- community
Checking that the SKILL.md file is still in the repository.
Use after an approved design and research brief to create a detailed, executable multi-step implementation plan before code changes.
Install
npx vskill@latest install minhtran3124/agent-harness/writing-plansSKILL.md
Writing Plans
Create specs/<slug>/PLAN.md for work that needs a plan. Announce this skill, then Read `rules/plan-format.md` and Read `rules/terminology.md` before authoring: both are path-scoped and a new plan does not load them automatically. From terminology.md, §3 is the load-bearing one — every Verify:, Done:, and Success Criterion must be decidable by running something, never by judging that output "is correct" or "looks right".
Required inputs
Read specs/<slug>/design.md and specs/<slug>/research-brief.md. Stop and tell the user when a required brief is missing. If the scope contains independent subsystems, propose separate plans.
Authoring flow
- Map created/changed files and their responsibilities. Follow existing patterns and give each
unit a clear boundary; do not introduce unrelated refactors.
- Write
## 3. Success Criteriafirst. Each observable behavior needs a re-runnable, pipe-free,
sub-60-second check and expected exit code, using the canonical schema from plan-format.md.
- For every new plan, add exact
## Global Constraints. Decompose tasks in canonical### Task
syntax. Give every task Files, imperative Action, automated Verify, measurable Done, mapped Criteria (SC-n), and an Interfaces line naming what it consumes/produces. Make actions test-first where applicable.
- Ensure same-wave tasks have no file overlap and no unresolved dependency. Use waves only for
genuinely independent work.
- When
design.mdhas a## What must be truesection, first write### Assumption coveragein
## 5. Risks, after the risk bullets with a blank line (start Risks with - none when there are no risks): the plan renderer cards only the first bullet list. Give each design A<n> one line: - A<n> → SC-n, - A<n> → Task x.y, - A<n> → accepted risk: <reason>, or - A<n> → retired by DC<n> when the design retired it; write - none (design has no assumptions) when that section is - none — <reason>. Check that every named SC-n and task exists in the plan. Then run python3 scripts/check_plan_contract.py specs/<slug>/PLAN.md, read references/review-loop.md, review the plan, and save it. Then run python3 skills/visual-planner/render_plan.py specs/<slug>/PLAN.md --summarize, which refreshes the At-a-glance block and writes the plain PLAN.html; use visual-planner only when the user asks for its graph-derived overlay.
Handoff
State the saved plan path. Before any code change, use using-git-worktrees unless already isolated, then invoke subagent-driven-development. Ask which execution mode only when the user has not stated one; both modes use the same gates.
Security
- Security scanPassed, 95 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/writing-plans/SKILL.md
Popularity
- Repository stars
- 7
- 7-day trend score
- 10
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/minhtran3124/agent-harness/writing-plans)