Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add HsuanYuLee/polaris --skill check-your-own-workgit clone --depth 1 https://github.com/HsuanYuLee/polarisWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/hsuanyulee/polaris/check-your-own-work)<a href="https://agentmods.dev/skills/hsuanyulee/polaris/check-your-own-work"><img src="https://agentmods.dev/badge/skills/hsuanyulee/polaris/check-your-own-work/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/hsuanyulee/polaris/check-your-own-work"><img src="https://agentmods.dev/badge/skills/hsuanyulee/polaris/check-your-own-work.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00254 | $0.03776 |
| Opus 5 | $0.00127 | $0.01888 |
| Sonnet 5 | $0.00051 | $0.00755 |
| Haiku 4.5 | $0.00025 | $0.00378 |
Grade A, and why
check-your-own-work scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 192 lines — stays where its author put it; the contents beside it link to each section on GitHub.
check-your-own-work — 交出去之前,先對一次自己寫的東西
這不是一道關卡。 它不回 PASS/FAIL,也不阻止任何後續動作。它產出一份 finding 清單, 而那份清單的價值完全來自它在同一輪裡被處置掉——一份沒有人動的報告,跟沒有報告一樣。
八問不是想出來的。前六問是從 829 則真人 review 意見逆推出來的六類反覆缺陷,而其中四類 完全不需要任何領域知識就避得掉:它們是「我沒有把自己剛寫的東西跟自己剛寫的宣稱對一次」, 不是「我不知道這個框架怎麼寫」。
第七問的出處不同,要分開說。 它來自 2026-08-27 一組十二則意見——那一批走了六輪 review, 而八則 SWE 通則裡有五則是同一個根因:拿掉一個機制,只補上它看得見的那一半。它不在 829 則 那個統計裡,所以不要把它讀成「第七類反覆缺陷」——它是一類,但樣本只有一批。
第八問的樣本更薄,只有一批十二則。 那一批歸出四個根因,其中三個前面七問都抓不到: 一個新加的測試檔沒有任何 CI 會跑、301 行另一種語言的測試設施在一個零先例的 repo 裡、 以及三支既有測試對唯一的行為改變全綠(第三個由第六問放寬管轄接手)。兩問共用同一次 搜尋,所以它們是一問不是兩問。同樣不要把它讀成一個統計結論。
先把材料撈出來
bash .claude/skills/check-your-own-work/scripts/collect-self-check-inputs.sh
bash .claude/skills/check-your-own-work/scripts/collect-self-check-inputs.sh --repo <path> --base <ref> --pr <number>
三個參數都可以不給:repo 預設是當下的目錄,base 自己去問 origin/HEAD(問不到就依序試
main/master/develop),PR 自己去問 gh。它唯讀,不對外寫入任何東西。
它逐問印出那一問需要的輸入,拿不到的那幾問指名說出為什麼拿不到,最後印
ANSWERABLE: n/8。這一行是整支腳本存在的理由:一份只答了兩問的自檢,讀起來跟答滿八問的
一模一樣,所以它要說出自己少了哪幾問。
答不出不是通過。 沒有 gh、沒有 PR、沒有上一輪意見、這個 repo 一份規範都沒有、diff
是空的——這五種都是常態,不是錯誤,但它們各自代表「這一問沒有答案」,不代表「這一問沒事」。
八問
一、我寫下的每一句宣稱,在 diff 裡都找得到對應的改動嗎
宣稱不只是 PR 描述。commit message、changeset、註解、docstring、型別宣告、變數名——每一句 描述性的話都是產出的一部分,不是旁白。它對讀的人承諾了某件事,而承諾與行為是分開演化的: 先寫下承諾,做的過程中改了做法,然後沒有回頭改那句話。
沒有東西會紅。編譯器不看,測試不看,只有下一個依它行事的人會踩到。
標本:四個 PR 各自宣稱有一組 harness 檔案,那些檔案不在 diff 裡;同時各帶一個宣告某個套件 要發版的 changeset,而那個套件一行都沒動。
二、我碰到的這些檔案類型,這個 repo 自己的規範說了什麼
從 repo 現場讀,不從記憶讀。 腳本會把它找到的規範檔逐份列出來——去讀它們。
兩件事讓「憑記憶」特別危險:
- 規範會翻面。 同一份檔案上個月說 A,這個月改成 B,而記得舊版本的人不會發現自己記錯了。
- 被違反的正好是已經寫下來的那些。 量到的第 2 類缺陷裡,reviewer 引用的就是這個 repo 自己的規範檔——引用它的是 reviewer,不是作者。
一份規範都找不到時,那不是「沒有規矩」,是「規矩沒有寫下來」——去讀鄰近的既有程式碼。
三、這個修法是不是一個 pattern
把這次的修法講成一句話,然後 grep 整棵樹找同型的地方,說出還有幾處、以及為什麼那幾處 不改。腳本印的「同目錄還有幾個同副檔名」是一個下界,不是答案。
「我只改了我看到的那一處」不是理由。標本:一個顯示問題的成因被作者寫進其中一個檔案的註解裡, 另外三處一模一樣的地方沒動。
四、我講的哪些 runtime 行為是實測的,哪些只是從源碼推的
源碼分析是假設,不是證據。 把這一輪講過的每一句 runtime 行為列出來,逐句標記。推的那 幾句要嘛去跑一次,要嘛在交出去的時候明講它是推的。
What ships with it
2 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 192 lines · 254 tokens per session scan A 88b3c4f4df57
check-your-own-work is a skill published in the GitHub repository HsuanYuLee/polaris (5 stars, last pushed yesterday), licensed MIT. It adds 254 tokens to every session and 3,776 once invoked, about $0.0013 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
resolve-pr-comments
Evaluate, fix, answer, and reply to GitHub pull request review comments and conversation comments. Handles both change requests (fix or skip) and reviewer questions (explain using reasoning recalled from past Claude Code transcripts). Use when the user asks to "resolve PR comments", "fix review comments", "address PR…
review-code
Review code for bugs, security vulnerabilities, API misuse, consistency issues, simplicity problems, or test coverage gaps by running internal reviews and a peer review in parallel and returning combined findings. Single-concern with a type argument, or full review with no argument. Use when the user asks to "review…
assess-technical-debt
Assess project-wide structural technical debt: complexity hotspots, deprecated API usage, duplication clusters, and architecture rot. Ranks findings by impact and refactor effort into a report at .turbo/technical-debt.md. Use when the user asks to "assess technical debt", "find technical debt", "review technical…
map-codebase
Deep architecture report that fans out parallel inspections across different aspects of the codebase (structure, tech stack, APIs, patterns, data flow, dependencies, testing) and synthesizes findings into a comprehensive document at .turbo/codebase-map.md and .turbo/codebase-map.html. Use when the user asks to "map…
peer-review
Run an independent peer review via Codex. Use when the user asks to "peer review", "peer review my code", "peer review my plan", "get a second opinion", or "independent review".
claude-print
Run a non-interactive Claude Code print-mode call from Codex. Use when the user asks to "claude print", "ask claude", "run claude", "consult claude", or when a Codex Turbo skill needs Claude as an independent peer reviewer.