plan
Plan a feature into a PRD with vertical-slice issues in docs/plans, in grill mode or trust mode. Runs only when the user types /plan.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-13
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- plan
- metadata
- { "opencode/autoinvoke": false }
- description
- Plan a feature into a PRD with vertical-slice issues in docs/plans, in grill mode or trust mode. Runs only when the user types /plan.
- disable-model-invocation
- true
02 资源文件 9
- 参考文档agents/openai.yaml96 B
- 其他includes.toml517 B
- 参考文档references/domain-modeling/ADR-FORMAT.md2.7 KB
- 参考文档references/domain-modeling/agents/openai.yaml101 B
- 参考文档references/domain-modeling/GLOSSARY-FORMAT.md2.2 KB
- 参考文档references/domain-modeling/SKILL.ref.md3.3 KB
- 参考文档references/matt-grill-with-docs.md247 B
- 参考文档references/matt-grilling.md2.0 KB
- 参考文档references/pstack-unslop.md5.7 KB
03 SKILL.md 正文
plan
Turn an idea into a PRD and a set of vertical slices.
Pick the mode
The first word of the argument picks the mode: /plan grill or /plan trust. The rest of the argument is the idea.
- Grill mode. Interview the user first, then write the plan. Use it when the idea is fuzzy, decisions are open, or the user wants you to challenge their thinking.
- Trust mode. Ask nothing. Build the plan from the conversation, the code, and the knowledge base. Use it when the context is already rich or the user wants speed.
With no mode word, recommend one mode in one line and wait. Base the choice on how much context exists. For example: "Context is thin, so grill mode fits. Say trust to skip the questions."
Knowledge base
Read the project's existing glossary. If the project has none, use docs/CONTEXT.md. Read the ADRs in docs/adr/ for each area the plan touches. Follow the decisions in those ADRs.
Keep both files current as decisions settle, in both modes:
- Follow
references/domain-modeling/SKILL.ref.mdto challenge fuzzy terms and to decide when a decision needs an ADR. - Take the entry formats from
references/domain-modeling/GLOSSARY-FORMAT.mdandreferences/domain-modeling/ADR-FORMAT.md. - Where that skill names
GLOSSARY.md, use that glossary file.builduses the same file. - Write each term and each ADR when it settles. Do not save them for the end.
Grill mode
Run the interview in references/matt-grilling.md. Apply the domain-modeling rules during the whole interview.
references/matt-grill-with-docs.md pairs those two skills. Where it says to call the Skill tool, read the two files above. Never load a skill of that name from another plugin, even when the session has one.
Look up facts in the code yourself. Put only decisions to the user. The interview ends when the frontier is empty or the user says "done". The settled decisions go into the PRD.
Trust mode
Work only from the conversation, the code, and the knowledge base. Do not ask. Write each assumption in the Decisions section. Use the glossary terms. Respect the ADRs. Put a real blocker in Open questions. Do not guess at it.
The PRD
Write it to docs/plans/NNNN-<slug>.md. NNNN is four digits with leading zeros, one more than the highest number in docs/plans/. The first plan in a repo is 0001. This sequence is separate from docs/adr/.
Use these sections:
- Problem. The problem from the user's side, not the system's.
- Solution. What changes for the user.
- User stories. A long numbered list: "As a <actor>, I want <feature>, so that <benefit>."
- Decisions. Modules touched, schema and API contracts, and settled trade-offs, in glossary terms. Mark each one as grilled or assumed. Use no file paths and no code. The one exception is a snippet that a prototype proved and that states a decision better than prose.
- Open questions. Each open point, so that none is lost.
- Out of scope. Say it explicitly.
Apply each rule in references/pstack-unslop.md to the PRD and the slices.
Slice
Cut the work into vertical slices. Each slice goes through every layer (schema, API, UI, tests) and can be demoed alone. Prefer many thin slices to a few thick ones.
Give each slice a title, the behavior to build (not the layers), acceptance criteria as checkboxes, and what blocks it.
In grill mode, show the slices and change them until the user approves. In trust mode, deliver them. The user edits after.
Publish
By default, the slices stay in the PRD as sections with checkboxes. If the repo has a GitHub remote and the user wants issues, file them with gh issue create. File blockers first, so each "Blocked by" names a real issue number.
Sources
- h00man, by h00man23 (https://github.com/h00mankind/workflow, Apache-2.0): the earlier
h00man:planskill. It gives the two modes, the knowledge base rules, the PRD sections, the slices, and the publish step. - matt, by Matt Pocock (https://github.com/mattpocock/skills, MIT): the
grillingskill, copied toreferences/matt-grilling.md, for the interview in rounds. Thegrill-with-docsskill, copied toreferences/matt-grill-with-docs.md, for the pairing of grilling with domain modeling. Thedomain-modelingskill, copied toreferences/domain-modeling/, for term challenges, ADR rules, and the entry formats. - pstack, by Lauren Tan (https://github.com/cursor/plugins/tree/main/pstack, ported at https://github.com/h00mankind/pstack, MIT): the
unslopskill, copied toreferences/pstack-unslop.md, for the numbered catalog of AI patterns to keep out of the PRD.
04 同仓库更多 Skills
teach
Teach the user a subject, in explain mode (one plain account of code, a change, or a concept, in chat) or course mode (a learning workspace with a mission, lessons, and records over many sessions). Runs only when the user types /teach.
council
Get a review or a second opinion on existing work from several strong models of different providers. Runs only when the user types /council.
variant
Builds multiple variants of a component you're working on and helps you iterate and pick one.
teardown
Creates technical illustrations of physical objects and of how things work, such as display pixels or interface layers, with coherent views, fine linework and part fills.
state-machine
Renders every state of a component you choose on a throwaway page, with mock data and a switcher, so you can work on each state.
interface-review
Reviews a branch, pull request or uncommitted change for the interface problems it introduced or regressed, across accessibility, layout, writing, typography, color and UI.