Skill v1.0.1
currentLLM-judged scan90/100+1 new
version: "1.0.1" name: oracle description: One second-opinion router for code, planning, security/deep audit, and creative critique. NOT for routine lookups (use grep/read/recall-memories) or fire-and-forget work (use dispatch). when_to_use: "oracle, second opinion, sanity check, ask oracle, planning oracle, plan critique, code critique, creative critique, security audit critique, red team critique, disagreement after reasoning" user-invocable: true
Oracle
One router for second opinions. User text wins: if the user explicitly asks for astra, sol, fable, creative, code, planning, or security routing, follow that route.
Dispatch
| Request | Route | |
|---|---|---|
| Code review, bug hunt, algorithm/design critique | fable subagent | |
| Planning critique, release plan, architecture plan | fable subagent | |
| Security, red-team, exploitability, deep audit | fable subagent | |
| Naming, prose, narrative, product copy, ideation | astra skill, high effort | |
| Ambiguous but touches code or operations | fable subagent | |
| Explicit "ask codex" / "use codex" | astra skill, high effort | |
| Explicit Astra or Sol request | astra skill; for Sol, its § Sol variant |
Fable Route
Launch a background agent with subagent_type: "fable" — agents/fable.md pins it at xhigh, so every fable route runs at xhigh; the opus fallback (agents/opus.md) runs at high. Include the goal, target files/dirs, and what to return. Frame adversarially:
Goal: <X>. Find the flaw in <code/design/plan>. Entry points: <files, symbols,error output>. Return findings only, with file:line or concrete trace.
Rules:
- NEVER set effort in the prompt text — ALWAYS rely on the agent file's pin.
- Never ask "does this look right?" Ask what breaks, what is missing, or why the
plan fails.
- Verify claims against the repo before acting.
- When fable cannot run (a usage limit), ALWAYS take the same route on
subagent_type: "opus" — NEVER skip the opinion.
Astra Route
Load the astra skill and follow its runbook. Use it for creative critique and explicit Codex or Astra requests. Keep the prompt adversarial and high-level; do not paste your full reasoning chain.
For an explicit Sol opinion, ALWAYS run astra § Sol (gpt-6.1-sol, ephemeral) under the same astra § Rules; NEVER answer a Sol request on the Astra model silently.
When the Astra CLI call cannot run (revoked auth, missing CLI), ALWAYS report the failure and take the Fable Route, on opus when fable cannot run either. For a missing pinned model, ALWAYS follow the named skill's stop-and-report rule; NEVER substitute an engine without the user's choice.
Output
Treat every oracle answer as advisory. Cite the finding when acting; discard wrong claims with a one-line reason.