- name
SKILL.md
技能檔案保留在發布者的儲存庫中。
在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add reticlehq/reticle --skill false-green-tests -a cursor -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add reticlehq/reticle --skill false-green-tests -a claude-code -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add reticlehq/reticle --skill false-green-tests -a codex -y在終端機中貼上複製的指令。它會將這個 SKILL.md 安裝到對應的代理中。
npx skills add reticlehq/reticle --skill false-green-tests -a grok -y在 Grok Bot 聊天中傳送複製的訊息。Bot 會從 GitHub 取得這個 SKILL.md,並加入你所有 Bot 共用的技能中。
Install this skill: https://github.com/reticlehq/reticle/tree/HEAD/skills/false-green-tests來自發布者
A green test is evidence about the test, not about the app. This skill separates the two by running the real app and comparing what it does against what the test claims.
It uses Reticle, which observes the running app from the inside: DOM, network, console, routing, and framework state. Not set up? Follow the reticle skill: it installs Reticle, asks the user before init changes the app's files, and proves one flow.
1. The assertion cannot fail. Read the test the user trusts. If it only asserts absence (no console error, no thrown exception, no rejected promise) it passes on a control wired to nothing. A dead button throws nothing, fires nothing, and changes nothing. Prove it in the app instead:
reticle_act_and_wait({ sessionId, ref, action: "click", until: { kind: "allOf", predicates: [
{ kind: "net", method: "POST", urlContains: "/api/...", status: 200 },
{ kind: "signal", name: "..." },
]}})
A verdict of no here, against a green suite, is the false green.
2. The test drove a mock and the app drives an API. Compare what the app actually requested with what the test stubbed:
reticle_observe({ action: "network", sessionId, since })
No request where the test asserted one means the suite verified a fixture. A stale client cache is the same failure with no request at all to look at, which is why registering TanStack Query matters: the cache is the only witness.
3. The UI moved and the state did not. The strongest false green, and invisible to any DOM or screenshot check:
reticle_look({ action: "state", sessionId, store, path })
A view rendering one value while the store holds another is a bug the render tree cannot show you. If this returns empty or hasCapabilities is false, no store was registered: say so, because every state check above is vacuous until it is.
4. Nobody read the response. A 2xx that the app never consumed shows as verified: "unknown" with verifiedReason: "outcome_unread". That is usually a real app bug, and a test asserting on the request alone would call it a pass.
5. Whole surfaces nobody exercised.
reticle_run({ tool: "reticle_verify", sessionId, args: { action: "crawl" } }) // click sweep, returns anomalies
reticle_run({ tool: "reticle_verify", sessionId, args: { action: "coverage" } }) // { total, exercised, untouched }
crawl exists because the obvious hand-rolled sweep (click each control, assert no console error) is itself shape 1.
A false green is confirmed when the app contradicts the test, not when you feel uneasy. Reticle names that case directly: verified: "no" with verifiedReason: "contradicted" means a channel observed something incompatible with what the UI claimed: a request that failed while the screen advanced, a signal disagreeing with the DOM, a field echoing a value nobody asked for.
unknown is not a false green and not a pass. It means Reticle could not tell. Report it as unknown.
For every false green you confirm, the test that missed it is still there and will miss it again. Rewrite its assertion to name a consequence the app must produce (a request with a status, a signal, a state path) rather than an absence. Never weaken a check to make a verdict green; that is how the false green got in.
Index of everything, one page at a time: curl https://docs.reticle.sh/llms.txt. Found a case Reticle could not see? reticle_session { action: "feedback" } with kind: "gap": that is the signal that decides what gets built.
Find out why the tests pass but the app is broken. Catches false greens: a green suite over a feature that does not work, a mocked API standing in for a real one, an assertion that holds no matter wha
---name: false-green-testsdescription: Find out why the tests pass but the app is broken. Catches false greens: a green suite over a feature that…---# skills/false-green-tests/SKILL.md總覽
Find out why the tests pass but the app is broken. Catches false greens: a green suite over a feature that does not work, a mocked API standing in for a real one, an assertion that holds no matter what the app does, a click handler wired to nothing. Use when the suite is green and the user says it is broken, when a test never fails, when coverage looks fine but bugs still ship, or before trusting a passing run you did not watch.
常見問題
從發布者下載 skills/false-green-tests/SKILL.md,或複製路徑後開啟儲存庫。PlugMCP 只收錄這項技能,不代管檔案。
發布者將它存放在 https://github.com/reticlehq/reticle。
Claude、Cursor、Codex 和 Grok Build 可以讀取你從發布者儲存庫加入的 SKILL.md。在聊天中傳送技能的 GitHub 資料夾連結,Grok Bot 就會安裝它。