菜单
Use when mining external JS/TS libraries, runtimes, papers, benchmark suites, bug corpora, or production systems for ideas that should become local tests, fuzzers, benchmarks, or evidence notes.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when adding ShapeShift Labs Evidence Kit to a JS/TS project, initializing the full tests/fuzzers/benchmarks/package-boundary/perf-wiki harness, or helping an AI agent establish maximum evidence coverage in another repository.
Use when creating, adapting, reviewing, or debugging JS/TS fuzzers, corpus fixtures, replay files, shrinkers, repro writers, or deterministic random test harnesses.
Use when adapting ShapeShift Labs Evidence Kit to a specific JS/TS project by deriving target-owned source lists, correctness contracts, corpus cases, fuzz generators, benchmark fixtures, source-pass plans, or evidence acceptance criteria from that project's own code, docs, tests, issues, or fetched sources.
Use for JS/TS performance-sensitive changes, benchmark additions, benchmark regressions, benchmark result JSON, or any claim that code is faster, smaller, cheaper, or not slower.
Use when changing JS/TS package exports, startup/import cost, package size, public subpaths, dependency direction, optional peers, tree-shaking, or package-boundary tests.
Use when creating or updating JS/TS performance evidence notes, iteration logs, research notes, generated docs/perf indexes, or searchable benchmark/result summaries.
| name | source-pass-harvester |
| description | Use when mining external JS/TS libraries, runtimes, papers, benchmark suites, bug corpora, or production systems for ideas that should become local tests, fuzzers, benchmarks, or evidence notes. |
Source passes turn external inspiration into local evidence. They are a fetch/cache/classify/transfer loop, not just agent-written notes.
npm run docs:perf:search -- <topic terms>
npm run research:list
Define the question: what weakness or workload are we trying to learn about? If the target project still lacks target-owned contracts, corpus seeds, or benchmark fixtures, use target-evidence-designer first.
Add or update a repeatable fetcher and source config:
npx evidence-kit add-source-fetcher --name <topic>
npm run research:fetch -- <topic>
Fetchers should cache external repos under research/repos/<topic>/, raw data or metadata under benchmarks/data/<topic>/, and write research/repos/<topic>/manifest.json.
Inspect fetched material for test practice, fuzzer structure, benchmark fixture shape, failure handling, package boundaries, startup budgets, and corpus curation.
Transfer ideas into local artifacts:
Keep runtime dependencies out unless the benchmark proves they are worth it.
Record rejected ideas explicitly.
Accept an idea only when it is:
Create or update a research/*-sources.md note and an iterations/*-source-pass.md note. Include source names, URLs, commits or hashes, local cache paths, manifest path, commands, accepted transfers, rejected transfers, and artifact paths.