[jasontopia]

Intent

把这份报告放到 learn-firstmate 下面合适的地方，作为参考资料。

"这份报告"指的是一份刚完成的调研报告（中文，2026-09-28），主题是 addyosmani/agent-skills 这套 engineering skills 对 FirstMate 有没有用。报告内容包括：agent-skills 的九个命令、六个阶段和 25 个 skill；按四层（编排、交付验证、工程纪律、产品定义）分析它和 FirstMate 的重叠、互补与冲突；FirstMate 的 worker 怎样项目级挑选安装这些 skill；与 Superpowers、Matt Pocock skills、Spec Kit、OpenSpec、Kiro、BMAD 等竞品的对比；以及结论：以 FirstMate + no-mistakes 为主干，项目级挑选安装约 12 个纪律类 skill，用 scout + Lavish 产出 SPEC。

What Changed

• Adds references/agent-skills-evaluation-2026-09-28.md, a Chinese research snapshot dated 2026-09-28 that asks whether addyosmani/agent-skills is useful for FirstMate. It covers the nine commands, six phases and 25 skills. It compares agent-skills with FirstMate across four layers (orchestration, delivery verification, engineering discipline, product definition) and marks where they overlap, complement or conflict. It also explains how crewmates can pick and install skills per project, compares competitors (Superpowers, Matt Pocock's skills, Spec Kit, OpenSpec, Kiro, BMAD) and gives the recommendation: FirstMate + no-mistakes as the backbone, about 12 discipline skills installed per project, and a SPEC produced with scout + Lavish.
• The report is pinned to specific commits (agent-skills 2686b62, FirstMate b3dbc67, no-mistakes v1.79.0), and its opening note says so. Its terms follow GLOSSARY.md, and content tied to one captain's personal setup was removed.
• Adds a "参考资料" section to README.md. It says reference material is a dated snapshot, not a chapter with an exercise, and links to the new report with a one-line summary.

Risk Assessment

✅ Low: The change is documentation only: it adds a reference report and a README index entry that match the stated intent. The only issue is that the report's claim of unified terminology is inaccurate, which is a content-quality concern, not a correctness or safety risk.

Testing

The baseline docs check already passed, and I re-ran it for evidence: it confirms the README link to the report and the report's link to GLOSSARY.md both resolve. I checked by grep and by reading that the report is in a sensible references/ location, is linked from README, covers every topic in the intent, uses the glossary terms the user chose to fix, and contains no captain-private paths. No screenshots: this is a Markdown-only repo with no rendered app or runtime, so there is no live surface to capture.

• Live validation: ⚠️ no-surface - 0 of 4 scenarios driven live against the product

Scenario	Result	Live	Evidence
Reader opens README, finds the 参考资料 section, and follows the link to the agent-skills report	⏸️ untested	no	Docs-only Markdown repo with no running product. Link integrity was checked statically only; rendering it on GitHub after push would be the live check.
Report covers every topic the intent names (commands/phases/25 skills, four layers, crewmate installation, competitors, conclusion)	⏸️ untested	no	Content check by reading the file. There is no runtime surface to drive.
Adversarial: report still uses non-glossary terms (worker, ship/scout 任务, no-mistakes 流水线, English-first captain) despite claiming glossary unification	⏸️ untested	no	Static text check on a docs-only change. There is no runtime surface.
Adversarial: report leaks captain-private local paths or data	⏸️ untested	no	Static text check on a docs-only change. There is no runtime surface.

Evidence: docs check output

checked 17 markdown file(s): no problems found

checked 17 markdown file(s): no problems found

- Outcome: ⚠️ 1 warning across 1 run (31.3s)

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

⚠️ **Review** - 1 info

• ⚠️ references/agent-skills-evaluation-2026-09-28.md:9 - The header note says terms were unified with GLOSSARY.md ("术语按 术语译法表 统一"), and AGENTS.md requires one rendering per term across the whole book. The report body still does not follow the glossary. It uses the untranslated "worker" about 36 times where the glossary says 船员 (lines 16, 28, 146, 162, 174-184, 217, 276 and others). It uses "worktree" where the glossary says 工作树 (lines 28, 144, 164, 181, 236, 283), "ship / scout" and "ship 任务 / scout 任务" where it says 交付任务 / 调研任务 (lines 32, 91-93, 190, 209, 273-276), and "no-mistakes 流水线" where it says 质检流水线 (lines 34, 181, 224). The concept table at lines 26-38 also puts English first ("captain（船长）"), against the glossary's Chinese-first convention. A reader who trusts the header gets inconsistent terms, and the claim in the note is wrong. Remedy: either apply the glossary across the report, or reword line 9 (and the README blurb) to say the report keeps its original terms. Choosing between these depends on the author's intent to keep the report 'as is', so this needs a decision.
• ℹ️ README.md:10 - The change adds a new top-level references/ directory and a README section for it. AGENTS.md ("What this repo is" / "Writing rules") only describes chapters/NN-slug.md and chapter rules, such as the plain style and every FirstMate claim traceable to the checkout. It does not say whether reference snapshots are exempt from those rules. Future agents may try to 'fix' the report to chapter standards, or may apply its commit-pinned line numbers as current fact. Consider adding one line to AGENTS.md stating what references/ is and which rules apply to it.

🔧 Fix applied.
1 info still open:

• ℹ️ README.md:10 - The change adds a new top-level references/ directory and a README section for it. AGENTS.md ("What this repo is" / "Writing rules") only describes chapters/NN-slug.md and chapter rules, such as the plain style and every FirstMate claim traceable to the checkout. It does not say whether reference snapshots are exempt from those rules. Future agents may try to 'fix' the report to chapter standards, or may apply its commit-pinned line numbers as current fact. Consider adding one line to AGENTS.md stating what references/ is and which rules apply to it.

⚠️ **Test** - 1 warning

• ⚠️ this change has no live-validatable surface; proceed without live validation? (0 of 4 scenarios were driven live against the product); Reader opens README, finds the 参考资料 section, and follows the link to the agent-skills report: Docs-only Markdown repo with no running product. Link integrity was checked statically only; rendering it on GitHub after push would be the live check.; Report covers every topic the intent names (commands/phases/25 skills, four layers, crewmate installation, competitors, conclusion): Content check by reading the file. There is no runtime surface to drive.; Adversarial: report still uses non-glossary terms (worker, ship/scout 任务, no-mistakes 流水线, English-first captain) despite claiming glossary unification: Static text check on a docs-only change. There is no runtime surface.; Adversarial: report leaks captain-private local paths or data: Static text check on a docs-only change. There is no runtime surface.
• Live validation: ⚠️ no-surface - 0 of 4 scenarios driven live against the product

Scenario	Result	Live	Evidence
Reader opens README, finds the 参考资料 section, and follows the link to the agent-skills report	⏸️ untested	no	Docs-only Markdown repo with no running product. Link integrity was checked statically only; rendering it on GitHub after push would be the live check.
Report covers every topic the intent names (commands/phases/25 skills, four layers, crewmate installation, competitors, conclusion)	⏸️ untested	no	Content check by reading the file. There is no runtime surface to drive.
Adversarial: report still uses non-glossary terms (worker, ship/scout 任务, no-mistakes 流水线, English-first captain) despite claiming glossary unification	⏸️ untested	no	Static text check on a docs-only change. There is no runtime surface.
Adversarial: report leaks captain-private local paths or data	⏸️ untested	no	Static text check on a docs-only change. There is no runtime surface.

• python3 scripts/check-docs.py
• python3 scripts/check-docs.py (link integrity incl. README -> references/ and references/ -> ../GLOSSARY.md; 17 files, no problems)
• git diff 28ef745 HEAD -- README.md to confirm the new 参考资料 section and its link
• grep of the report for the old term renderings (worker, ship/scout 任务, no-mistakes 流水线, captain（) - 0 hits each; worktree hits reviewed by hand
• grep of the report for private paths (/Users/, caishen) - none found
• Heading outline and keyword check that the report contains every topic the intent names (9 commands/6 phases, 25 skills, four layers, crewmate installation, Superpowers/Pocock/Spec Kit/OpenSpec/Kiro/BMAD, ~12 skills, scout + Lavish SPEC)

✅ **Document** - passed

✅ No issues found.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.