Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/HsuanYuLee/polarisnpx agentmods add skills/hsuanyulee/polaris/verify-acWrote 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/verify-ac)<a href="https://agentmods.dev/skills/hsuanyulee/polaris/verify-ac"><img src="https://agentmods.dev/badge/skills/hsuanyulee/polaris/verify-ac/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/verify-ac"><img src="https://agentmods.dev/badge/skills/hsuanyulee/polaris/verify-ac.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 | $0.00178 | $0.06855 |
| Opus 5 | $0.00089 | $0.03427 |
| Sonnet 5 | $0.00036 | $0.01371 |
| Haiku 4.5 | $0.00018 | $0.00685 |
Grade A, and why
verify-ac 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 5d 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 — 349 lines — stays where its author put it; the contents beside it link to each section on GitHub.
verify-ac — 第二關:執行 oracle
輸出有兩部分,權力不一樣:機械判定會擋(由 exit code 承載,不需要讀者同意); 判斷報告不擋(帶引用的意見,由人裁)。把兩者混在一起,關卡就會開始擋一些沒人能精確 定義的東西,然後大家學會繞過它。
機械判定:三件事
# 1. assertion 沒被動過。這一步同時查兩件事:校驗值與內文對得上,且內文與 git 歷史一致。
# 對不上就停——這時審查根本還沒開始,因為成功的定義變了。
bash .claude/skills/verify-ac/scripts/frozen-assertion-fence.sh verify {issue}/index.md
# 2. 手上這條量測命令是登錄過的。換過而沒帶紅過證據的命令不被承認。
bash .claude/skills/verify-ac/scripts/record-measurement-change.sh verify \
--ledger {issue}/.spine/measurement-ledger.json \
--assertion-id A-P1 --command '<cmd>'
# 3. 跑量測。同時看 exit code 與正向證據——exit 0 而沒有正向證據不是通過。
bash .claude/skills/verify-ac/scripts/run-hardened-oracle.sh --command '<cmd>' \
--require-tool rg --expect-evidence '<真的量到東西的痕跡>' \
--evidence-out {issue}/.spine/evidence/<assertion-id>.json
任一項不成立就是非 PASS,沒有討論空間。
一條命令被好幾條 assertion 共用時,用 --assertion 分組,一趟就產出全部證據:
bash .claude/skills/verify-ac/scripts/run-hardened-oracle.sh --command '<cmd>' \
--assertion A-P1 --expect-evidence '<A-P1 自己的痕跡>' --evidence-out {issue}/.spine/evidence/A-P1.json \
--assertion A-P2 --expect-evidence '<A-P2 自己的痕跡>' --evidence-out {issue}/.spine/evidence/A-P2.json
命令只跑一次,每一組在同一份輸出上各自判、各自寫——不是把同一個判定複製 N 份,那會讓 「這條 assertion 真的被檢查過」變成假的。一組沒有自己的輸出路徑、或兩組指到同一個檔案,它會停。
run-hardened-oracle.sh 會先探工具能力再釘住、要求命令產出證明自己量到東西的輸出、
並原樣保留 stderr 與 exit code。
釘住的意思是那條命令跑在一組固定的系統目錄上,所以沒有用 --require-tool 宣告的工具,
在它底下就是不存在。 那不是理論上的:2026-08-30 有兩支 selftest 用 rg,在一般 shell
底下全綠,在這條命令底下直接 command not found。而它紅的樣子跟「斷言真的沒過」離場碼
相同——分得開它們的只有輸出裡那一行說著缺了什麼。這是因為工具會說謊:PATH 上較早的 shim、靜默跳過的測試、
被吞成 generic timeout 的錯誤,三者都能讓一個空的執行看起來像綠的。
命令與探針都不准把位置抄下來
一張單住在哪一格是它的狀態的投影,會被重算。 所以登錄下來的命令、以及探針解 repo 根 的寫法,兩邊都不准把「現在在哪」寫成字面值——寫下去的那一刻它就是一個會在下一次重算之後 失效的指標,而失效的樣子是「開不到檔」,讀起來像交付壞了。
命令這一邊,執行當下才問位置:
bash "$(bash .claude/skills/driving-work-to-done/scripts/spine-loop-state.sh find <單的目錄名>)/probes/probe.sh"
後面不要接 | tail -1。find 命中不是剛好一個的時候會回非 0(3 = 多於一個,4 = 一個
都沒有),而那個離場碼會被 tail 吃掉,於是兩行路徑被當成一條用。record-measurement-change.sh
在登錄的時候會擋下把單自己的目錄名連著斜線抄進去的命令,三種寫法都算:展開後的絕對路徑、
以 $HOME 開頭的、相對於另一棵樹的。
What ships with it
23 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.
- scripts/check-spine-legacy-layers.sh 5.7 KB runs code
- scripts/enumerate-spine-inventory.sh 9.7 KB runs code
- scripts/frozen-assertion-fence.sh 18 KB runs code
- scripts/lib/assertion_verdicts.py 53 KB runs code
- scripts/lib/evidence_report.py 8.6 KB runs code
- scripts/lib/probe-root.sh 2.6 KB runs code
- scripts/record-delivery-intent.sh 25 KB runs code
- scripts/record-measurement-change.sh 22 KB runs code
- scripts/render-evidence-report.sh 5.8 KB runs code
- scripts/report-assertions.sh 3.0 KB runs code
- scripts/resolve-evidence-publish.sh 5.3 KB runs code
- scripts/run-hardened-oracle.sh 20 KB runs code
- scripts/selftests/check-spine-legacy-layers-selftest.sh 8.0 KB runs code
- scripts/selftests/enumerate-spine-inventory-selftest.sh 9.2 KB runs code
- scripts/selftests/evidence-report-selftest.sh 14 KB runs code
- scripts/selftests/frozen-assertion-fence-selftest.sh 14 KB runs code
- scripts/selftests/no-orphans-selftest.sh 6.8 KB runs code
- scripts/selftests/record-delivery-intent-selftest.sh 22 KB runs code
- scripts/selftests/record-measurement-change-selftest.sh 8.7 KB runs code
- scripts/selftests/report-assertions-selftest.sh 13 KB runs code
- scripts/selftests/run-hardened-oracle-groups-selftest.sh 9.1 KB runs code
- scripts/selftests/run-hardened-oracle-selftest.sh 8.9 KB runs code
- scripts/selftests/standalone-selftest.sh 3.7 KB runs code
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.
- 5d ago Changed · +5 lines 3cc32f07be26
- 10d ago First seen · 344 lines · 178 tokens per session scan A 75e62a58ab39
verify-ac is a skill published in the GitHub repository HsuanYuLee/polaris (5 stars, last pushed today), licensed MIT. It adds 178 tokens to every session and 6,855 once invoked, about $0.0009 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
api-tester
A tool for creating and checking API tests from the real API contract and implementation. An API is the agreed way that software sends requests and receives responses.
test-generator
A test-writing helper that creates unit or integration tests from the code's actual behavior and contracts. Unit tests check small pieces of code, while integration tests check how real components work together.
smoke-test
Launch the app and hands-on verify that it works by interacting with it. Falls back to an existing integration test suite when there is no interactive surface in scope. Use when the user asks to "smoke test", "test it manually", "verify it works", "try it out", "run a smoke test", "check it in the browser", or "does…
exploratory-test
Execute multi-level exploratory testing of the app covering basic functionality, complex operations, adversarial testing, and cross-cutting scenarios, plus usability observations through a UX lens reported separately from defects. Deeper than /smoke-test. Use when the user asks to "exploratory test", "test…
create-test-plan
Analyze what changed and generate a structured test plan at .turbo/test-plans/ .md covering four escalating levels: basic functionality, complex operations, adversarial testing, and cross-cutting scenarios. Use when the user asks to "create a test plan", "plan tests", "what should I test", "generate test scenarios"…
quick-finalize
Close out a change without the deep review loop: stage, simplify code and docs, run the project's checks, smoke test, update the changelog, self-improve, and ship. Use when the user asks to "quick finalize", "quickly finalize", "finalize quickly", "light finalize", "wrap this up quickly", "close this out without the…