Audit-and-fix method for one user setting area across every repo. Use when a setting seems ignored, duplicated or confusing, on 'audit the X settings' or 'why doesn't my X setting apply', or as a lane of the settings sweep. NOT for adding a new setting (use build-sub-feature).
インストールする前に、エージェントに与えられる指示の中身を確認できます。
Arman, 2026-09-25, after two wrong answers about voice: "The core problem is no one gives enough fucks to go get the facts." The voice area had five pickers in four stores; three of them were read by nothing, one claimed to apply "everywhere", none played a sample, and the Chrome extension used a hardcoded voice. Worked example — read it once: for-Arman hand-over 2026-09-25 · voice-census.md (deleted in d0e885c98).
Every step ends on a check. The report is the census table plus the fixes, never prose.
Start from where the capability actually happens, in every repo (matrx-frontend, aidream
incl. mobile/, matrx-extend, matrx-local, matrx-sandbox, packages). Grep the doers, not the
settings: for voice that was speak(, websockets, speechSynthesis, TTS endpoints, realtime
clients, TwiML. Group call sites into user-visible experiences. One row each:
experience · route/surface · audience · engine/provider · vocabulary (the allowed values) · value comes from (file:line) · user can change it where · entry UI (has a menu?)
Audience is consumer (the person using AI Matrx for themselves) or builder (a value chosen for something they build that speaks to other people — a podcast host, an agent's voice, a workflow step). Never let one control serve both.
Done when: two independent methods (grep of doers + enumerate routes/clients) add no new row.
For each settings control in the area: the key it writes → grep every reader of that key in every repo. No reader that runs (a reader nothing renders counts as none) = dead control.
Done when: every control row says read by <file:line> or DEAD.
For each experience: which store it reads, what it falls back to, whether that fallback is a hardcoded constant, and when it resolves (a one-shot read before the organization loads silently takes the fallback — make it reactive).
Can one value mean the same thing to every consumer? If the engines' vocabularies differ, there is no single setting — one row per real vocabulary, stated honestly ("each engine has its own voices"). Faking one value across engines is the defect this skill exists to stop.
| Class | Looks like |
|---|---|
| Dead control | saves a value nothing reads |
| Duplicate control | two screens edit one key, one without the good UI |
| Rival store | two keys for one idea; consumers split between them |
| Ignored setting | a consumer (often another repo/client) reads a constant instead |
| Lying copy | "used everywhere", "app-wide", a label naming the wrong scope |
| Hidden default | an override screen that does not show what applies when unset |
| No preview | a sensory choice (voice, sound, theme, font) you cannot hear/see first |
| No door | the place the setting acts has no path to the row that governs it |
| Consumer/builder blend | a personal preference rewriting what a builder made, or vice versa |
| Timing race | resolved once before org/session loads, so the fallback wins |
platform.feature_knob, organization
→ user → device), rendered by the ONE editor (KnobFieldControl); give it a preview when sensory.SettingDoor → the exact row). A lone icon gets the door in
its menu, never a second icon.platform.knob_snapshot).As the test account in the browser: change the value, hear/see it at each consumer, reload, follow each door and land on the highlighted row. A picker that renders but never plays is not verified.
The census table (step 1), the control table (step 2), the defects found by class, and what changed with file paths — then commit per repo and add the area to the sweep register.
まだレビューはありません。使ってみた感想をお寄せください。
概要と使いどころ
Admin debug system wiring for the floating AdminIndicator's live debug data and Copy Full Context. Use when adding debug visibility to a route or feature, wiring useDebugContext or the debug panel, capturing console errors, or enabling the copy-context workflow for a page.
日本語の概要は準備中です。原文の説明を表示しています。
Compact two-icon Copy / Copy-for-AI controls (components/agent-copy). Use when adding copy buttons to a row, card, list, or record; merging duplicate Copy/JSON/AI controls; continuing the copy rollout; or writing a Copy-for-AI payload. NOT for markdown content actions (use rich-document-actions).
日本語の概要は準備中です。原文の説明を表示しています。
Disclosing a surface's existing fixed AI jobs in the shell's top Agents menu. Use when a page, panel, overlay, or window already runs a mandate behind a button, assist, automatic action, or mode; when the agent-disclosure guard names a file; or during a surface check.
日本語の概要は準備中です。原文の説明を表示しています。
The watch-fix-rerun method for making an existing platform agent or automated process efficient and correct: baseline its ledger, run one unit yourself, fix the class behind every wasted call, rerun, record. Use when asked to improve, optimize, watch, or 'make efficient' an agent, a sandbox session, a sync job, or any recurring automated process.
日本語の概要は準備中です。原文の説明を表示しています。
Redux state for agent execution and firing agent shortcuts. Use when editing features/agents/redux/, building agent UI, creating a conversation, touching assembleRequest or NDJSON stream state, wiring agent-state selectors, adding a per-conversation capability, or triggering a shortcut from a button, menu, or code (useShortcutTrigger, launchAgentExecution).
日本語の概要は準備中です。原文の説明を表示しています。
The provision for a call site: the exhaustive menu of values that place in the code can realistically produce. Use before creating or fixing a mandate or agent that reasons about more than its own input, when deciding whether an agent can answer at all with what it is sent, or when its output 'looks right' but nobody checked it could be. NOT for building the agent (use create-agent).
日本語の概要は準備中です。原文の説明を表示しています。