本文へ移動
cccskills
無料GitHub で公開

agent-disclosure

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.

インストール方法を見る

含まれるファイル(2)

  • SKILL.md8.9 KB
  • agents/openai.yaml377 B

SKILL.md(原文)

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。

Agent disclosure — register existing fixed jobs in the top menu

The non-negotiable boundary

🚨 DISCLOSURE MUST NEVER CHANGE THE SURFACE'S VISIBLE CONTENT.

Do not add an agent chip, badge, card, row, label, roster, callout, explanatory copy, icon, toolbar item, or section to the page, panel, dialog, overlay, or window. Do not move or resize existing content to make room for disclosure. Do not create a new agent integration merely so a surface can be "complete."

The disclosure destination is the shell's existing top Agents menu only. The job must already exist in the product surface. This skill records that pre-existing fixed job so the user can inspect its mandate without leaving the page. It is not authority to modify page content or product behavior.

One owner-approved exception (Arman, 2026-09-25): the small Intelligence icon (IntelligenceIndicator, features/mandates/feature-intelligence/FEATURE.md) may sit beside a control that runs a job. It is a door to where the job is managed, not a disclosure roster — never add anything else under its name.

There is no inline disclosure component. PageAgents is forbidden and deleted. If a task or old document tells you to render it, that instruction is stale and must not be followed.

Cross-repo source of truth: ../../../common-docs/systems/mandates/CLIENT-SURFACES.md § THE DISCLOSURE LAW. Frontend detail: features/surfaces/FEATURE.md.

Decide whether a roster exists

Fixed surface job — disclose in the top menu

A fixed job is an AI action the surface itself owns: for example, a specific mandate behind "Assign topics" or a recurring agent that performs the same job for this page. Confirm the integration already exists before registering it.

  • Static job: declare a manifest agentRole carrying its mandateKey.
  • Runtime-selected fixed job: call useDeclaredSurfaceMandates from the existing action-bearing component. This call has no UI.
  • Open the mandate through useOpenMandateWindow; never link away to a mandate route from a working surface.

Agent-native and resident agents — explicit isolation

Chat, Agent Builder, runners, tests, and window-panel residents may host an outside helper that works on the page's data. They are not blanket binding exclusions. The page's native subject agent and any permanent resident helper must never acquire ad-hoc surface values merely because it is placed on, shown in, or bound to that surface. Their engineered state and context remain theirs.

Each invocation follows an approved context mode: none, named_inputs, or approved_surface_context; the latter may use explicit mappings or a user-approved task-appropriate page snapshot. A documented integration contract or explicit user choice is approval provenance and persists until its scope changes; it does not require a new prompt for every run. Placement or a saved binding does not approve a mode or any value. Record the agent/job, surface, purpose, mode, allowed value names, approval provenance, and evidence before treating a helper context handoff as complete. This documentation sets the required contract and evidence; it claims neither runtime enforcement nor a unified tracker implementation.

  • A native subject launch receives only its explicit engineered inputs. It never receives ambient or ad-hoc surface context; helper approval cannot create an exception.
  • A permanent resident, including a window-panel helper, has no automatic inheritance. Its deliberate job contract may use named inputs or approved surface context when it works on page data.
  • An outside helper may use any page, including Chat or Agent Builder, under its recorded integration contract. Binding availability and a data handoff are separate decisions; saving a binding does not require or grant context.
  • When the approved purpose is agent definition work, the outside helper may edit every user-editable definition field, enter runtime variables and human input, and run the test. The contract records the actual installed targets; it does not pretend an unavailable target or runtime enforcement exists.
  • Disclosure still names only an existing fixed job in the top Agents menu. It never adds visible agent content or invents an agent.

Find the work

pnpm check:agent-disclosure

The guard scans UI surfaces that run mandates. It also fails loudly if the forbidden PageAgents import or JSX pattern returns anywhere. Execution machinery is excluded because it runs on behalf of a surface.

Before changing anything, inventory the surface's existing buttons, assists, automatic actions, tabs, and modes. For every candidate, prove which fixed mandate it already runs. Never infer a job from an agent-looking icon or from the existence of an agent picker.

Register a static fixed job

In the surface manifest:

agentRoles: [
  {
    name: "topic_assigner",
    label: "Topic assigner",
    description: "Places this surface's keywords onto the Offering tree.",
    kind: "single",
    defaultAgentId: null,
    mandateKey: "seo.topic_assigner",
    autoRun: "never",
    sortOrder: 100,
  },
],
  • Use a mandate key, never a raw agent UUID.
  • does/description says what the job does here in plain product language.
  • mandateKey and a non-null defaultAgentId are mutually exclusive.
  • Do not add a role when the surface merely lets the user choose an agent.

Register a runtime-selected fixed job

Use this only when live state determines which already-existing fixed job the surface runs:

useDeclaredSurfaceMandates(
  engine.agents.map((agent) => ({
    mandateKey: agent.mandateKey,
    does: agent.does,
    surfaceName,
  })),
);

useDeclaredSurfaceMandates populates the top Agents menu and renders nothing. Mount it in the existing action-bearing component. Do not add a wrapper or a visible sibling. Do not use it for arbitrary picker choices.

Open the mandate in place

const openMandate = useOpenMandateWindow();
openMandate({
  initialMandateKey,
  mandateKeys,
  surfaceName,
  initialView: "yours",
});

The existing menu row is the door. A link from a working surface to /mandates or /administration/mandates is a regression.

Decision table

SituationVerdict
Any surface, including Chat or Agent Builder, has an existing outside helperIt may bind there; its data delivery follows its recorded approved integration contract.
The launch is the page's native subject agentGive it only its engineered inputs; never an ambient or ad-hoc page snapshot.
The launch is a permanent resident helperUse none, named inputs, or approved surface context only under its deliberate documented contract.
An existing fixed job runs on a surfaceRegister it in the top menu only; this is separate from context approval.
Runtime mode selects one of several existing fixed jobsRegister the live fixed job without rendering UI.
On an eligible ordinary product surface, service/thunk/tool machinery runs a mandateRegister on the calling surface, not in machinery.
No AI job exists yetDo nothing; disclosure cannot invent one.
Agent uses a raw UUID with no mandateFix hardcoded-agent architecture first; do not invent a key.

Verify live

  1. Capture the surface before and after. The page body's visible content, spacing, controls, and layout are unchanged by disclosure.
  2. For a fixed job, open the existing top Agents menu and confirm the job appears exactly once. Open it and confirm the mandate window appears over the page.
  3. For each helper invocation, verify the saved record and a run: allowed names arrive for approved_surface_context; no ad-hoc surface values reach a native or resident agent whose mode is none.
  4. Confirm no PageAgents import, JSX, agent chip, or disclosure-only page content exists.
  5. Run pnpm check:agent-disclosure, focused tests, and pnpm type-check.

Sweep discipline

  • One feature area per commit; stage only owned files in the shared checkout.
  • Never add visible UI to reduce the guard count.
  • Never register an agent the surface does not already run.
  • Never add an exemption merely because a finding is inconvenient.
  • Update the feature's FEATURE.md change log in the same commit.

Where this runs from

  • Standalone: /agent-disclosure or an undisclosed-surface sweep.
  • Live UI work: live-ui-iteration routes here when the surface already has a fixed AI job.
  • Certification: surface-check S5.
  • Codex and Claude share this exact skill because .agents/skills points to .claude/skills; agents/openai.yaml supplies the Codex interface.

レビュー

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。

同じリポジトリのスキル

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。

armanisadeghi/ai-matrx32026年10月13日 更新

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。

armanisadeghi/ai-matrx32026年10月13日 更新

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。

armanisadeghi/ai-matrx32026年10月13日 更新

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。

armanisadeghi/ai-matrx32026年10月13日 更新

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。

armanisadeghi/ai-matrx32026年10月13日 更新

The agent.review_queue register for work Arman must see or test in the UI. Use at the end of any task that produced something reviewable, at the start of a task to check prior feedback, and when reviewing, claiming, routing, or repairing a queue row.

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。

armanisadeghi/ai-matrx32026年10月13日 更新

armanisadeghi のスキルをすべて見る

このスキルの問題を報告する