メニュー
Use when delegating GitHub repository reconnaissance to the ghx sidecar service: ask repo questions in plain English and receive compact evidence reports.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/gkoreli/ghx --skill ghx-reconコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Use when asked to audit the ghx product from a product-manager's perspective — to find issues, gaps, north-star misalignment, PMF / market / competitive questions, adjacent-idea proposals, and agent-experience problems. Runs the product through MULTIPLE adversarial PM personalities across strategic and agent-native scopes and surfaces (code, docs, features, evals, live manual runs, the strategy docs themselves). Agentic-first: ghx's primary consumer is an AI agent, so classic PM advice is re-priced before it is applied. This is a PRODUCT/strategy audit, not a code-quality or security audit.
Use when creating a NEW skill (or a substantial reference / strategy / design doc) that deserves real rigor — built via wide delegated research, an authored draft, and iterative adversarial review rounds, with the orchestrator as the authoritative final judge. This is the heavyweight path for artifacts that will be reused and must be high quality; do NOT use it for trivial or one-line skills, mechanical config, or a quick doc — just write those directly.
Use to run a SCOPED, multi-persona, partly-adversarial audit of ghx's architecture, codebase, and tech debt — Go architecture/design-pattern quality, encapsulation, domain modules, core/reusable-capability boundaries (the Agent Sidecar Framework as reusable infra; the shared core between evals and the main ghx product), and tech debt in specific modules. Background personas (several adversarial) each write one ADR-style artifact into docs/audits/; a distillation artifact converges them; the orchestrator makes the final judgment and a governing ADR. TRIGGER-GATED, not periodic. This is a CODE/ARCHITECTURE audit — for product/PMF/strategy use `product-audit`; for a quick one-file review just read it.
| name | ghx-recon |
| description | Use when delegating GitHub repository reconnaissance to the ghx sidecar service: ask repo questions in plain English and receive compact evidence reports. |
| version | 1.0.0 |
| author | ghx contributors |
| license | MIT |
| metadata | {"hermes":{"tags":["github","code-exploration","sidecar","repository-recon"],"related_skills":[]}} |
ghx is a specialist agent that explores GitHub for you: ask a repo question in plain English and it returns a compact evidence report. Never explore a remote repo step-by-step yourself — delegate the whole question and let ghx do the reading, mapping, and searching under the hood.
Two entry points, same service:
recon tool — recon(question, repo?).ghx sidecar ask "<question>".State the goal and name the repo when you know it — "How does hono implement middleware chaining, and which files define it?" beats "run grep for middleware". Omit the repo for discovery questions — "which repos or libraries do X" — and ghx sweeps GitHub, verifies the top candidates by reading them, and ranks them. One investigation per question; follow up rather than bundling.
Every answer ends with an artifacts pointer (session dir + trace id) — the on-disk audit trail behind the report. Follow-ups stay on the right investigation automatically; you never manage sessions. A fresh question takes tens of seconds; follow-ups are faster.