honesty-kits

Agent Receipt Kit | Checks an AI agent's claims against its allowed actions, cited evidence, and supplied observations.

Trust Core | TypeScript scoring helpers for contributor and source signals with caller-set weights.

Cost Governor Kit | TypeScript tools for estimating AI call costs, setting spending limits, and reconciling estimated with confirmed usage.

Mutation Invariance Kit | TypeScript tests that check whether selected inputs change a decision or ranking in specified scenarios.

Payout Invariance Kit | Tests and source checks for whether selected payout inputs change rankings in specified scenarios.

Claims Registry Kit | Tracks product claims, supporting evidence, and when each claim needs review.

Small, dependency-free checks for the claims your AI product makes. Try all 12 kits in your browser.

Freshness Kit | TypeScript helpers that label content fresh, aging, or stale using supplied review dates and thresholds.

Provenance Kit | TypeScript helpers for labeling claims as verified, modeled, or editorial, checking related wording, and rendering them.

Advice Ledger Kit | Compare a recommendation with what happened later, or show where a model and person disagree; refuses to grade thin evidence.

Corroboration Kit | TypeScript rules that compare caller-labeled sources and evidence coverage to flag when support is limited.

Honesty MCP | One server that gives AI assistants tools for checking citations, evidence, claims, and decisions. Requires the sibling kit repositories.

Claude radical honesty

Grounding Kit | Offline first-pass checks of citation markers and text overlap with supplied evidence.

Audit Chain Kit | TypeScript tools to link audit records and detect changes against a supplied chain head.