소스 정보
- 저장소
- whiteguo233/OpenBiliClaw
- 최근 소스 활동
- 2026년 7월 10일 08:43
- 감지된 SKILL.md 언어
- 영어
- 스타
- 3,223
- 포크
- 160
설치 방법
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
소스 파일 검토
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
메뉴
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/whiteguo233/OpenBiliClaw --skill writing-specs명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | writing-specs |
| description | Use when authoring a spec or implementation plan, or planning multi-task work. |
Produce an authoritative design spec and an executable implementation plan before changing code. Resolve every placeholder against the live repository.
Create both documents with the same dated slug:
docs/plans/YYYY-MM-DD-<slug>-spec.mddocs/plans/YYYY-MM-DD-<slug>-plan.mdCommit the pair as docs: add <feature> spec and plan before implementation. Do not mix code
into this docs-first commit.
# <Feature> Spec — <measurable outcome>
**Created:** YYYY-MM-DD
**Scope:** <affected behavior, modules, and integrations>
**Out of scope:** <explicit non-goals>
## Goal
State the current cost or failure, quantified target outcomes, and the exact commands that
verify them.
## Design invariants (MUST hold in every phase)
1. **<Invariant name>:** <falsifiable rule and verification surface>.
2. **<Invariant name>:** <falsifiable rule and verification surface>.
## Current diagnosis
### D1. <Observed problem>
Record evidence at `path/to/file.py:line`; distinguish confirmed facts from hypotheses.
### D2. <Observed problem>
Record the caller, data flow, failure mode, and existing test coverage with `file:line` evidence.
## Priority classification
| Phase | Content | Tier | Why |
| --- | --- | --- | --- |
| 0 | <gate or prerequisite> | **MUST** | <risk it controls> |
| 1 | <main change> | RECOMMENDED | <measured leverage> |
List dependencies, then group work into **Wave A**, **Wave B**, and later waves by risk and
dependency. State what can ship independently and where work may safely stop.
## Phase designs
### Phase 0 — <name>
Define interfaces, algorithms, error behavior, tests, rollout, and numeric acceptance gates.
## Expected impact
| Lever | Measured effect |
| --- | --- |
| <phase> | <quantified result> |
## Documentation obligations
List every module doc, changelog, architecture diagram, CLI/config reference, installer doc,
and README surface triggered by the design.
Every invariant must be testable, every diagnosis must cite live code, and every claimed improvement must have a reproduction command.
# <Feature> — Implementation Plan
> **For Claude:** REQUIRED SUB-SKILL: superpowers:executing-plans (execute this plan task-by-task).
> **Spec:** [`YYYY-MM-DD-<slug>-spec.md`](./YYYY-MM-DD-<slug>-spec.md)
> **Status:** <revision and review state>
> **Execution order:** <dependency-ordered tasks and Wave grouping>
> **Tech:** <runtime, interpreter, focused tests, lint, format, and type-check commands>
**Invariants that MUST hold — re-read before each task:**
- <Restate the spec's invariant without weakening it.>
- <Restate the spec's invariant without weakening it.>
### Task N: <single deliverable>
**Files:** Add/modify/test exact paths.
Consumes: . Produces: .
[ ] Write one focused failing test for .
[ ] Run and confirm FAIL for the intended missing behavior.
[ ] Add the minimal implementation needed for that test.
[ ] Rerun and confirm PASS with no warnings.
[ ] Run the touched regression tests, lint/format checks, and MyPy command.
Numeric gate: .
Reproduce with ; record the result in the PR.
State the production/shadow/canary observation, commands, owner, duration, and rollback trigger.
Keep the handoff line and checkbox steps verbatim so superpowers/GSD execution conventions remain discoverable. Restate every spec invariant at the plan top and require the executor to re-read them before each task.
For a model-visible profile reduction, a concrete gate is: admission flip rate ≤ 3% and
Spearman rank correlation ≥ 0.95 on at least 100 aligned candidates, verified by
scripts/run_profile_diet_ab.py. The command, dataset provenance, baseline commit, and observed
values belong in the spec and the task acceptance block—not only in a PR comment.
Use
docs/plans/2026-07-05-llm-token-diet-spec.md
as a reference for quantified invariants, D1..Dn diagnosis, risk Waves, phase designs, impact,
and documentation obligations.