# Plan > Create an implementation plan and save it to docs/specs/_/plan.md, gated by a required constitution check. - **Type:** Skill - **Install:** `agentstack add skill-konh-claudetools-plan` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [KonH](https://agentstack.voostack.com/s/konh) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [KonH](https://github.com/KonH) - **Source:** https://github.com/KonH/ClaudeTools/tree/main/plugins/k/skills/plan ## Install ```sh agentstack add skill-konh-claudetools-plan ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About Create a plan for the requested task and save it to `docs/specs/_/plan.md`. ## Doc paths Default convention: `docs/specs/` for spec/plan folders, `docs/constitution.md` for project principles. Check the project's `CLAUDE.md` for overriding locations before falling back to these defaults. ## Constitution Gate (required) A plan may not be finalized without a constitution. Before anything else: 1. Check whether `docs/constitution.md` exists. 2. If it does not: stop, tell the user no constitution was found, and ask whether they want one created now — and what principles it should capture. Do not invent principles on their behalf. Only continue once a constitution exists (pre-existing or just created with their input). Once a constitution exists, the architect sub-agent reads it before finalizing the plan and appends a **Constitution Check** section to the plan: - Either: `No conflicts found — plan aligns with all principles.` - Or: a numbered list of principles the plan would violate, each with a one-sentence proposed resolution The orchestrator surfaces any violations to the user before presenting the final plan. If there are violations, the user must confirm each resolution before the plan is written. Do not write the plan file if unresolved violations exist. ## Orchestration Spawn an **architect sub-agent** (general-purpose) to design and write the plan. Brief it with: - The user's task description - Relevant project rules (from `CLAUDE.md` and any project rules directory) - The contents of `docs/constitution.md` - Existing spec/plan folders in `docs/specs/` (list them to resolve an existing dated spec identifier) - The output path and format rules below The architect writes the plan file directly. You (orchestrator) then: 1. Present the plan contents to the user 2. Collect feedback and re-brief the architect if changes are needed (iterate until approved) 3. Run the `plan-review` skill as a final check — present any concerns one by one and ask the user to approve each fix 4. Stop and wait for the user to run the `implement` skill ## Spec Detection If `$ARGUMENTS` contains a spec folder name or index, or if a `docs/specs/_*` folder exists with no `plan.md` yet: - Read `docs/specs/_/spec.md` - Brief the architect with the spec contents - Include a **Spec** section at the top of the plan (verbatim summary of intent and acceptance criteria from the spec) If no spec folder is found (a purely technical task — migration, refactor, infra), create `docs/specs/_/plan.md` using the current local timestamp, just without a `spec.md` in that folder. ## Plan File Rules - The containing spec folder uses `YY_MM_DD_HH_name`, where the timestamp is the local creation time and `name` is a short kebab-case description (e.g. `26_07_18_14_map-prototype`) - Structure: goal, approach, steps — keep it concise - If the plan touches source code, include a **Tests** section covering what unit/integration tests should be added or updated - Do NOT make any code, asset, or file changes — only write the plan document - End every plan with the line: `Use the implement skill to start working on the plan or request changes.` ## Step Block Structure Split steps into two sections: **Section 1 — Agent Steps** (Claude performs autonomously via file edits and tools): - Use markdown checkboxes: `- [ ] **Step title** — concise description` **Section 2 — User Steps** (requires manual interaction Claude cannot perform — an external editor/tool, visual inspection, hardware, etc.): - Use numbered headings (`### N. Title`) with body text — no checkboxes ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [KonH](https://github.com/KonH) - **Source:** [KonH/ClaudeTools](https://github.com/KonH/ClaudeTools) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-konh-claudetools-plan - Seller: https://agentstack.voostack.com/s/konh - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.