本文へ移動
cccskills
無料GitHub で公開

no-dead-ends

Recipe for THE DOOR LAW (every record the UI names opens; every detected problem ships its fix) and THE INVENTORY LAW. Use before building or fixing a surface that names a record or flags a problem (cell, row, dialog, badge, warning, id, count), or on 'no link to X', 'why can't I click this', 'dead end', EntityRef, peek. NOT for page chrome (use core-route-headers).

インストール方法を見る

含まれるファイル(1)

  • SKILL.md6.7 KB

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。

no-dead-ends — every identity is a door, every capability is on the table

Read the doctrine first: ../common-docs/policies/no-dead-ends.md. It is canonical and cross-repo. This skill is the frontend mechanics: which primitive to reach for, how to wire it, and how to prove it. Out of scope: page-level chrome (use the core-route-headers skill) and choosing WHAT data a surface shows — this skill makes what is shown openable.

The two laws, one paragraph

If the UI names a thing that has an id in our database, the user must be able to reach it — THE DOOR LAW. And you may not build the surface before you know what the platform already gives it — THE INVENTORY LAW. A surface that prints "Deep Web Research Agent" as a <span> fails both: the agent has a route, an icon, a peek, an action registry, a lineage index, and a sync window, and the surface used none of them.

Step 1 — the inventory pass (before the first line)

For every entity the surface names, run these and write the results into your summary:

# Does it have a canonical route + icon?  → getEntityInfo(token).hrefFor
grep -n "hrefFor" features/scopes/registry/entityRegistry.ts

# Does it have a peek?
cat features/organizations/peek/kinds-list.ts

# Does it have an action registry / row-actions hook?
ls features/*/browse/*ActionRegistry.tsx components/official/item/

# Does it have an overlay opener or a window panel?
ls features/overlays/openers/ | grep -i <entity>
ls features/window-panels/windows/<entity>/

"Found nothing" is only acceptable when it names the queries you ran.

Step 2 — put the doors on

Use EntityRef (components/official/entity-ref/EntityRef.tsx). It is the Door Law made importable — name → route, plus new-tab, plus peek, all resolved from the registries. Never hand-roll a name link.

import { EntityRef } from "@/components/official/entity-ref/EntityRef";

<EntityRef token="agent" id={row.agentId} name={row.agentName} />

// Same record lives in two shells (system vs personal agents): override href.
<EntityRef token="agent" id={id} name={name} href={agentHref(id, agentType)} />

// Surface-specific extra doors go in `extraActions` — never forked into a copy.
<EntityRef token="note" id={id} name={title} extraActions={<OpenInWindowButton …/>} />

Safe inside clickable rows — every control stops propagation. Worked reference: features/mandates/admin/MandatesConsole.tsx (Agent column, Pin column, Health column, and the drawer's identity card).

Missing a door? Fix the registry, not the call site:

MissingFix
Routeadd hrefFor to the token in features/scopes/registry/entityRegistry.ts
Peekadd the kind to features/organizations/peek/registry.ts and kinds-list.ts (a dev-time guard screams if they drift)
Windowregister an opener under features/overlays/openers/ (see the overlay-system skill)

Never import features/organizations/peek/registry.ts for the availability check — it statically imports 19 peek components. Import hasPeek from kinds-list.ts and render the one dynamic edge, <ResourcePeekHost>.

Step 3 — the corollaries (where the real damage lives)

  1. Render every relationship you can resolve, with its own door. For agents the lineage is free: selectAgentLineageIndex / selectAgentLineage (../aidream/apps/shared/chat/src/agents/redux/agent-definition/selectors.ts) derive parent / children / systemTwin from the slice — zero extra queries. The twin can be a PARENT (personal copy of a system agent) or a CHILD (system agent promoted from a personal one); handle both. Cold surface with no slice? fetchLinkedCounterpart (thunks.ts) round-trips.

  2. Ship the fix beside the complaint. "NOT a system agent" must come with the rebind button and the twin's link, not a scolding badge.

  3. State the verdict, not the timestamp. "Last synced Apr 29" is not an answer. Say identical / what differs / link to the diff.

  4. Never render an id you can't open, and never report green for data you couldn't read — an unresolvable reference is its own loud state (unresolved pin in the mandates console), never ok.

  5. A count is a door. 3 overrides reaches those overrides.

  6. A SENTENCE is a door too. Our servers write refusals and notes for people, and those sentences name records by raw id — "resolved system agent 8f0bbfc2-… breaks the mandate contract". Printed as {message} that is a dead end with extra steps: the reader is told which record is wrong and then made to hand-copy a uuid into a URL bar. Print server prose through TextWithDoors (components/official/entity-ref/TextWithDoors.tsx) instead — it keeps every character verbatim and turns the ids into doors:

    import { TextWithDoors } from "@/components/official/entity-ref/TextWithDoors";
    
    <p className="text-destructive">
      <TextWithDoors text={refusal} defaultToken="agent" />
    </p>
    

    The token comes from the sentence's own words ("… agent <id> …"); pass defaultToken only when the SURFACE knows what its sentences are about. An id whose type cannot be established stays plain text — a link to the wrong record reads as a fact and is a lie. ServerNotes already renders through it, so every door's notes block inherits this for free.

Step 4 — prove it

  • Browser: every named record shows Open / new-tab / peek. read_page and confirm the link "Open X" / button "Quick look at X" / link "Open X in a new tab" nodes exist per row.
  • Data logic (lineage, twin resolution, health): a unit test, because the interesting rows are usually invisible to the test account under RLS — ../aidream/apps/shared/chat/src/agents/redux/agent-definition/__tests__/lineage-selectors.test.ts is the pattern. Never claim a path works because it "should".
  • Say plainly in your summary what you verified in a browser vs. by test.

Anti-patterns — fix on sight

  • <span>{record.name}</span> where hrefFor exists.
  • A red badge naming a problem with no action next to it.
  • A drawer that opens on a picker instead of showing what the user has.
  • A local <Dialog> preview when a peek is registered for that kind.
  • A hand-rolled row-action list beside an existing action registry.
  • A bare UUID cell.

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

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。

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

armanisadeghi のスキルをすべて見る

このスキルの問題を報告する