epistemic-protocols

The user vaguely recalls something discussed before but cannot name it — one session, or a line of work, topic, or settled concept across several: find it in past records to recognize.

This skill should be used when choi names a core protocol whose Lean contract is to be re-derived, or invokes /rederive. Runs the per-protocol flow: gather the pre-understanding, derive the protocol's epistemic solution and fuse it with choi's horizon, then read the current contract against it, through sketch, edit, dogfood, review, merge, the chart close and a retrospective on this skill. Project-local contributor tooling.

Scoped empirical validation. Decomposes a target↔surrogate equivalence claim into facets, bounds a test space, captures evidence inside it, and carries the untested complement forward.

Release for this repo — CalVer tag + narrative prepend.

Convergence-paced review-resolve loop over a change and its governing surfaces. Verifies each finding against the codebase and the base it is measured from, then re-reviews until each is disposed of.

Context about to be acted on may no longer hold — stale, weakly sourced, or contradicted: challenge each claim the action leans on, and the user judges what stands before acting.

Work-unit triage for GitHub issues. Groups raw issues, fuses each group with the AGENTS.md northstar, and externalizes each routed unit to a substrate record a collaborator session is pointed at.

This skill should be used when the user asks to "verify protocols", "check consistency before commit", "validate definitions", "run pre-commit checks", "verify soundness", or wants to ensure epistemic protocol quality. Invoke explicitly with /verify for pre-commit validation.

Use when the user asks to "check white bear", "audit prohibitions", "find negative framing", or invokes /white-bear. Read-only audit: unnecessary competing-target mentions in LLM-facing prose.

Use when the user asks to "check zero-shot", "audit few-shot anchoring", "find example anchoring", or invokes /zero-shot. Read-only audit of LLM-facing prose: principle over anchoring examples.