agentsclimarketplace

Oracle

Skill hammer-mt/oracle

Consult Claude Fable as a persistent, high-judgment advisory partner for planning and execution. Use when the user explicitly asks to invoke Oracle, consult Claude Fable, or get an external second opinion; says a problem is important, high-stakes, unusually hard, or likely to be difficult; when Codex is stuck, uncertain, or repeatedly failing; or before a consequential decision where an independent read on meaning, tradeoffs, blind spots, or sequencing would improve the result. Do not use for routine work that Codex can confidently complete directly.From its SKILL.md

Install
npx -y skills add hammer-mt/oracle

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.5 KB, 632 tokens by cl100k_base, as published. Nobody here has run it

Oracle

Use Claude Fable as a parallel judgment process. Keep one Claude session paired with the current Codex task, return to it with new evidence, and let Codex remain responsible for verification, writing, and execution.

Consult Oracle

  1. Do enough first-party work to identify the real decision. Inspect relevant artifacts and separate facts from guesses.
  2. Send a compact decision brief containing:
    • objective and why it matters;
    • constraints and irreversible choices;
    • evidence, attempts, and observed failures;
    • the decision or uncertainty Oracle should resolve;
    • only the new context since the prior consultation.
  3. Run the bundled wrapper from the active working directory:
python3 <skill-directory>/scripts/oracle.py ask <<'EOF'
Objective: ...
Why it matters: ...
Constraints: ...
Evidence and attempts: ...
Decision needed: ...
New since our last exchange: ...
EOF

The wrapper uses CODEX_THREAD_ID by default. Outside Codex, set ORACLE_THREAD_KEY or pass --thread-key <stable-conversation-id>. Never invent a different key for a later turn in the same task.

  1. Evaluate the advice against primary evidence. Oracle is an advisor, not an authority. Correct factual errors and reject suggestions that violate the user's intent or the repository's constraints.
  2. Incorporate the useful judgment into the plan or implementation in Codex's own clear voice. Do not paste Oracle's raw memo to the user unless the user asks for it.

Maintain the Paired Session

  • Let the wrapper create the first Fable session and persist its UUID. On every later consultation, use ask again; it resumes the stored session automatically.
  • Treat follow-ups as deltas. State what changed, what failed, and which decision is now live. Do not resend the full history that Fable already has.
  • Reconsult after material new evidence, a failed approach, a changed constraint, or a new decision fork. Do not ping Oracle merely to endorse a settled choice.
  • Inspect the pairing with python3 <skill-directory>/scripts/oracle.py status.
  • Never reset merely to get a different answer. If the stored Claude session is unavailable or its context is irreparably wrong, explain the continuity loss and use reset --force only with user approval.

Preserve Safety and Context Quality

  • Remove secrets, credentials, private user data, and irrelevant logs from the brief.
  • Give Oracle file paths only when their contents are safe to disclose to the locally authenticated Claude account.
  • Keep Oracle read-only. The wrapper starts Claude in safe mode, supplies a strict empty MCP configuration, and exposes only Read, Glob, and Grep with non-interactive permission handling.
  • Keep the stable system prompt, fable model alias, and xhigh effort default unless the user explicitly requests a change.
  • If Claude is missing, unauthenticated, or returns an error, report that plainly and continue with Codex's best judgment when safe; do not silently create a replacement session.

What ships with it: 4 files

14.3 KB alongside SKILL.md, 1 of them executable

agents/

scripts/

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.