hermes-labs-ai/hermeneutic/hermeneutic-gate

Hermeneutic gate

Use the installed hermeneutic CLI as a cheap, zero-LLM pre-flight check for assistant drafts. It catches fixed English surface patterns associated with completion overclaiming, relayed authority, unhedged certainty, scope expansion, and unsupported quality adjectives.

Workflow

1. Identify the exact draft that is about to be sent.
2. Run hermeneutic gate with the draft on standard input, for example:

``bash printf '%s' "$DRAFT" | hermeneutic gate ``

1. Inspect the printed verdict as well as the exit code. A low-severity

RISK is advisory and exits 0; medium- and high-severity RISK exits 1.

1. If the draft is held, add the missing evidence, hedge the claim, or remove

the unverifiable text before sending it.

The gate is advisory and does not edit the draft or call an LLM. Do not treat a pass as proof that a draft is correct, complete, safe, or free of novel drift. It checks English surface patterns only and does not replace human review.

Use this skill especially when the draft contains numeric claims, says that work is done or shipped, uses words such as “every” or “always”, or summarizes work completed by another agent.