fable-advisor

# fable-advisor Execute a task in the main agent (the **executor**) with a persistent **Fable 5 advisor** subagent overseeing it. The advisor plans and advises; the executor builds. A **separate fresh Fable 5 reviewer** — never the advisor — issues the binding pre-commit verdict, so the plan's author never grades its own implementation. When the task is a **GitHub issue**, the executor runs `work-on-issue`'s build-and-ship pipeline (gates, worktree, red → green verification, staging discipline, PR conventions, guardrail table) with the advisor and reviewer layered around it — `work-on-issue` stays the single source of truth for that pipeline, so any future hardening of it applies here too. **Issue-less prose tasks** take a lighter standalone path. ## Input The user provides one of: - A GitHub issue reference alone — full URL, `#<N>`, bare `<N>`, or `owner/repo#N`. **This is a complete input**: the task is the issue itself, derived from its title and body exactly as `work-on-issue` does — no accompanying prose description is needed. The plan and the final review verdict are also posted as issue comments. - A task in prose ("fable-advisor migrate X to Y") — the issue-less form. - Both — prose alongside an issue reference; the prose adds constraints, the issue defines the task. - If none of these is obvious from the invocation or the conversation, ask what to execute before dispatching. ## Model check This skill assumes the session model is NOT Fable 5 — the point is cheap...