AGENT SKILL

ulw-plan(omp)

Plan complex or ambiguous work before implementation. Use for ulw-plan(omp), plan mode, interview requests, architecture decisions, multi-module work, five-or-more-step work, vague briefs, or requests to break work down into one decision-complete plan.

License 许可
MIT
所属仓库
jhun-kim/omp-lazy
最近更新
2026-07-31
Stars
★ 0
Forks
0

01 前言 Front Matter

name
ulw-plan(omp)
description
Plan complex or ambiguous work before implementation. Use for ulw-plan(omp), plan mode, interview requests, architecture decisions, multi-module work, five-or-more-step work, vague briefs, or requests to break work down into one decision-complete plan.

02 资源文件 4

03 SKILL.md 正文

ulw-plan(omp)

Act as the omp-lazy-planner. Produce one decision-complete plan that a downstream worker can execute without another interview.

Plan mode is sticky. Stay in planning mode for the whole task. Treat requests to build, fix, or start as requests to plan. Read and analyze product files, but write only .omo/drafts/*.md and .omo/plans/*.md. Never implement or edit product code. Approval authorizes plan creation only.

Do not activate, create, or mutate native Goal state. If native Plan mode conflicts with an already-active Goal mode, report the conflict and stop without changing either mode.

Route intent

Ground the request in repository and primary-source evidence, then announce one route, the assigned tier (FAST, STANDARD, or DEEP per the tier classification in the full workflow), and whether review is required:

  • Intent: CLEAR when the outcome is known and only genuine owner decisions remain.
  • Intent: UNCLEAR when the outcome is open-ended and best-practice defaults must define it.
  • Route an explicit interview request to CLEAR even when the brief is fuzzy.
  • Break a genuine tie toward CLEAR and ask exactly one question.

Read exactly one routing reference plus the shared workflow:

Treat "high accuracy", "deep review", and equivalent language as review-required: true; do not use it to choose CLEAR or UNCLEAR.

Persist before asking

Run the shipped scaffold directly after routing:

node <skill-root>/scripts/scaffold-plan.mjs <slug> --clear|--unclear --draft

Record findings, decisions, assumptions, scope, tier, and status: awaiting-approval in the durable draft. Present the brief once and wait for explicit user approval. After approval, set status: approved and approval: explicit-user-approval, then run:

node <skill-root>/scripts/scaffold-plan.mjs <slug> --plan

Never add the approval marker from the original planning request. A later explicit reply must approve the recorded approach. On resume, read the draft and continue from its persisted status; do not reroute from memory.

Finish the plan

Use OMP task dispatch for read-only exploration and the namespaced planning agents. Record actual returned agent identities; requested labels are not identities. Append todos beneath the scaffold marker without rewriting emitted headers. Fill the human TL;DR last.

Run the tier-gated review sequence from the full workflow. Deliver the completed plan and review receipts, then stop. Never begin implementation.

04 同仓库更多 Skills