[self-care:agent-discoverability] Fix stale MCP wording and missing CAO skill routes for agent discovery
メンテナーはふだん 1 日以内に返信
まだ誰も着手していません。
評価
- 難易度
- 2/5
- 見積もり時間
- 1〜3時間
- 初心者へのやさしさ
- 62/100
- issue の種類
- ドキュメント
- 明瞭さ
- おおむね明確
- 活発さ
- 活発
- 技術スタック
- github-actions, go
- 領域
- cli, documentation
調査の方向性
指定された4か所から始めてください。docs/agent-analysis.md の「Node snapshot MCP server」の段落(183行目付近)、skills/cao-cli/SKILL.md の Targeted references の箇条書き(101行目付近)、skills/analyze-cao/SKILL.md の Shell procedure の手順2と Data contract セクション、そして手順4の鮮度に関する記述です。編集前に、internal/cli/native_mcp.go のツール一覧と、internal/cli/native_download.go のキャッシュパスの動作を確認してください。リンクが解決され、Node MCP に関する記述が残っておらず、npm run docs:build と node --test tests/unit/agent-docs-validation.test.mjs が成功すれば完了です。
索引モデルが issue の本文から書いたものです。
説明
Summary: The agent-discoverability audit of canonical CAO routes found no clean pass across its seven tasks. Source review at revision 185fd262b shows four routing defects and two gaps: stale "Node" MCP wording, a skill entry that promises cache-path content the referenced file lacks, no ingestion command in the analysis skill, and no link from the analysis skill to schema authority. Freshness semantics and the debug route are also undocumented for agents.
Action: A maintainer should assign this issue to Copilot, or apply the four documentation fixes below. Acceptance: all canonical links resolve, no Node MCP wording remains, and re-running the seven-task audit routes every task without repository search.
Agent prompt
In githubnext/gh-cao, make documentation-only routing fixes. Do not change code, policy, query semantics, schema, or workflow lock files.
- docs/agent-analysis.md, the paragraph beginning "The Node snapshot MCP server exposes exactly these two tools": replace "Node snapshot MCP server" with the Go
gh cao mcpserver. Keep the "exactly these two tools, in that order" claim only after verifying the registered tool list in internal/cli/native_mcp.go. - skills/cao-cli/SKILL.md, the Targeted references bullet for workflow-runtime.md says "cache paths, fallback behavior". Either correct that description to what the file covers, or add the cache-path content to skills/cao-cli/references/workflow-runtime.md after verifying it in internal/cli/native_download.go.
- skills/analyze-cao/SKILL.md, Shell procedure step 2 says an operator "must ingest" the artifact without a command. Add
cao-dashboard ingest --source .cao(as documented in docs/cao-cli.md and docs/dashboard-data-ingestion.md). In the Data contract section, link specs/dashboard-data.md as schema authority and specs/dashboard-gh-aw-jsonl-mapping.md for the activity shard mapping, replacing the unlinked phrase "the activity shard schema". - skills/analyze-cao/SKILL.md, step 4 tells agents to "establish freshness" without a stale threshold. Either link where stale is defined, or state that no threshold exists and that agents must compare as-of against the analysis window.
Optional: add a debug-cao row to the Common routes table in skills/cao-cli/SKILL.md, pointing deployment and campaign failure diagnosis to skills/debug-cao/SKILL.md.
Validate with npm run docs:build and node --test tests/unit/agent-docs-validation.test.mjs. Keep the change set to these skill and doc files.
Critical findings
docs/agent-analysis.md:183calls the MCP server "Node snapshot", butdocs/agent-analysis.md:245says the former Node MCP is removed. The live endpoint is the Gogh cao mcpcommand.skills/cao-cli/SKILL.md:101saysreferences/workflow-runtime.mdcovers "cache paths". That file contains no cache-path content.skills/analyze-cao/SKILL.mdrequires an operator ingest step with no command. Thecao-dashboard ingest --source DIRECTORYcommand appears only indocs/cao-cli.mdanddocs/dashboard-data-ingestion.md.skills/analyze-cao/SKILL.mdnames "the activity shard schema" and "the dashboard data specifications" without links. The authoritative files arespecs/dashboard-data.mdandspecs/dashboard-gh-aw-jsonl-mapping.md.- No stale threshold was found in the canonical docs searched. Agents are told to establish freshness with no rule.
debug-caois linked only from operator docs, not from the agent entry points.
Task table
| Task | Success | Entry point discovered | Skill used | Precomputed artifact used | MCP used | PostgreSQL used | UI scraping attempted | Repository search required | Navigation hops | Evidence |
|---|---|---|---|---|---|---|---|---|---|---|
| 1. Disabled campaigns | Partial | analyze-cao, then agent-analysis | analyze-cao | No | No | No | No | No | 3 | Route found. No snapshot or service in this run, so evidence is unavailable. |
| 2. Snapshot freshness | Fail | analyze-cao step 2, cao-cli | analyze-cao, cao-cli | No | No | No | No | Yes | 3 | Skill says to establish freshness. No stale threshold in the canonical docs searched. |
| 3. Named campaign failure | Partial | Not from analyze-cao or agent-analysis. Found in docs/operations.md | debug-cao (via operator docs) | No | No | No | No | Yes | 4 | debug-cao exists but has no agent-entry link. |
| 4. Recent failures via catalog | Fail | agent-analysis, "without a shell" | none | No | No. cao_catalog not exposed | No | No | Yes | 2 | Stale "Node snapshot MCP" text contradicts the removal note. |
| 5. Historical analysis | Fail | analyze-cao step 2 | analyze-cao | No | No | No | No | Yes | 4 | Ingest step has no command. The command lives in cao-cli.md and dashboard-data-ingestion.md. |
| 6. Schema authority | Partial | cao-cli guardrails, then specs/dashboard-data.md | cao-cli | No | No | No | No | Yes | 3 | analyze-cao names the shard schema without a link. |
| 7. Activity download and default path | Partial | analyze-cao step 2 | analyze-cao, cao-cli | No | No | No | No | Yes | 3 | The artifact is JSONL with no database file. Default .cao is in the flag help and docs/cao-cli.md, not in the skills. |
Evidence and limits
- Checkout revision 185fd262b. Published Pages URLs were not fetched; findings come from repository source.
gh cao --versionwas denied by the sandbox, so no CLI query ran.cao_catalogandcao_querywere not exposed to this run. MCP calls: 0. HTTP fetches: 0.- No PostgreSQL or operational snapshot access. No UI scraping.
- Repository search was needed for tasks 2 through 7 because the canonical entry points lack those links. Recorded as a defect.
cao: githubnext/gh-cao, correlation: 43637313167-4046
Generated by SelfCare / Agent Discoverability · copilot · routed: haiku55 xhigh · ~16 AIC · ⌖ ~2 AIC · ⊞ 8.7K · ◷
- expires on Nov 28, 2026, 8:41 PM UTC
- 主要言語
- JavaScript
- スター
- 19
- フォーク
- 4
- 平均マージ
- 4時間
- マージ済み PR(30日)
- 708
環境構築
- Dockerfile・Docker Compose ファイルなし
- プルリクエストのテンプレートなし
- コントリビューションガイドを読む
はじめの一歩
- issue を最後まで読み、次にプロジェクトのコントリビューションガイドを読みます。
- 着手することを issue にコメントします — 二人が同じ作業をするのを防げます。
- リポジトリをフォークし、ブランチを切って変更します。
- issue 番号を参照したプルリクエストを送ります。
githubnext/gh-cao のほかの issue
-
dependabot
難易度 2/5 1〜3時間 初心者へのやさしさ 68/100
githubnext/gh-cao#20776 · コメント 1 件 ·
メンテナーはふだん 1 日以内に返信
-
難易度 2/5 1〜3時間 初心者へのやさしさ 66/100
githubnext/gh-cao#20705 ·
メンテナーはふだん 1 日以内に返信
-
repo-assist repo-assist:maintenance
難易度 1/5 1〜3時間 初心者へのやさしさ 75/100
githubnext/gh-cao#20704 ·
メンテナーはふだん 1 日以内に返信
-
[uk-ai-advisory:operational-resilience] UK AI advisory: operational resilience for githubnext/gh-caoオープンuk-ai-advisory uk-ai-advisory:operational-resilience
難易度 2/5 1〜3時間 初心者へのやさしさ 68/100
githubnext/gh-cao#20669 ·
メンテナーはふだん 1 日以内に返信
-
self-care self-care:documentation-discoverability
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
githubnext/gh-cao#20667 ·
メンテナーはふだん 1 日以内に返信
githubnext/gh-cao の issue をすべて見る
似ている issue
-
framework/gatsby help wanted kind/bug
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
メンテナーはふだん 1 日以内に返信
-
good first issue help wanted javascript
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
dlq/synthpopcan#58 ·
メンテナーはふだん 2 日以内に返信
-
cli: `execCargoMetadata` decodes stdout per chunk, corrupting non-ASCII paths across pipe boundariesオープン
難易度 2/5 1〜3時間 初心者へのやさしさ 68/100
メンテナーはふだん 1 日以内に返信
-
難易度 1/5 1時間未満 初心者へのやさしさ 68/100
osmlab/name-suggestion-index#13104 ·
メンテナーはふだん 1 日以内に返信
-
bug triage
難易度 2/5 1〜3時間 初心者へのやさしさ 72/100
fjaeckel/seriesoftubes#376 ·
メンテナーはふだん 1 日以内に返信