Run the full on-track check on a conversation, project, or keyword the owner names: find the sessions, extract exactly what he typed, dispatch independent judges who go to the live product as him and report what was missed, plan the remedy, and return one page — what he will see, what is missing, the plan. Use when told to check whether work is on track, to audit a project that has drifted, on 'run the on-track check on X', or when a long campaign is about to freeze, hand off, or claim done.
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。
The owner names a thing — a conversation, a project, a phrase — and wants to know whether the
work is still what he asked for, what was missed, and what it would take to fix. You own the
whole run and return one page. Mechanics of each step live in user-ground-truth; the concept
behind all of it is reality is the referee: his words are
reality, every plan and register is a model, and the judges exist to let reality contradict the
model — so nothing you already believe goes into their briefs.
Sessions live under ~/.claude/projects/<cwd-with-slashes-as-dashes>/*.jsonl. The desktop app's
session ids do not name these files; his words are the key:
python3 /Users/armanisadeghi/code/common-docs/meta/scripts/transcript_user_messages.py \
~/.claude/projects/-Users-armanisadeghi-code*/*.jsonl --list --grep "<a phrase only he would have typed>" [--since YYYY-MM-DD]
One line per matching session — file, dates, message count, opening words. Pick from that; if he named a conversation title, the desktop session list gives its date and folder, and the opening words confirm the file. Never open a transcript to find one.
python3 .../transcript_user_messages.py <matching files...> -o <ground-truth>.md [--since ...]
Then a quick lane (Sonnet) with only that file and his summarization instruction verbatim
(user-ground-truth §2). Commit the file. Read it yourself once — it is short, and everything
you say to him later is measured against it.
One or more fresh lanes (standard; more than one only when the work spans clearly different
surfaces, each judge owning one). Each gets: the ground-truth file, the summary, and the work's
own claims by path (register, handoff, status, plan). The brief carries the situation and the
result wanted — where he said it should appear, what it should do, what he said mattered most —
and asks for what he asked for and cannot see, what was built somewhere else or beside something
that already existed, and what was reported done that is not there. It does not carry your
suspicions; withhold them and use them as the test of each report. A judge must go to the live
product as him and to the live data; a judge that only compares documents to documents has not
judged.
One fresh standard lane with his words, the summary, and the judges' reports: the remedy as a
plan in his terms — what changes where, what gets folded into what already exists, what gets
deleted (one path, never a twin), what he will see when it is done, and the order. It goes to
the product and the code too; it does not take the judges' word any more than they took the
builder's.
Lead with what he will see today doing what he described, then what is missing, then the plan, then three offers, numbered, in plain sentences:
Nothing is fixed, sent or scheduled until he picks. Counts, lane names and run ids stay out of the page; the judges' and planner's reports are linked for anyone who wants them.
Briefing the judges with what you think is wrong · a judge that never opens the product · summarizing the summary instead of his words · running it on a clock instead of at his ask or a decision point · sending him the reports instead of the page.
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。