Changelog · public since May 2026

What we shipped, when.

Reverse chronological. Older than what's listed here? Email team@agentlint.net.

  1. Harness-doc cross-checking

    Audits now fetch the canonical doc pages from Anthropic, Cursor, Aider, and OpenAI at audit time and cross-check your hooks and skills against the live spec. Catches things like a SessionStart hook emitting additionalContext when the spec is additional_context — a silent harness drop that nothing else surfaces.

  2. Bigger bundle, more findings, no truncation

    Bundle budget bumped from 100KB to 200KB; reordered so rule files, hooks, and harness specs land before depth-1 cross-references. With the model's reasoning trace also bumped from 8K to 16K tokens, audits no longer truncate JSON mid-finding.

  3. Severity-prefixed PR comments

    PR comments now use P0 / P1 / nit severity prefixes, GitHub committable suggestion blocks, and stay quiet on clean PRs. Review-body summary hides P1 and nit findings under a fold so the blocker is always the headline.

  4. Quoted text required on every model finding

    Every model-generated finding now requires a verbatim quote from the bundle. Speculation ("this could be improved") gets dropped at runtime — if there's no quote, there's no finding. Deterministic findings (broken pointers, missing files) don't quote because there's nothing to quote — they cite the dead reference instead.

  5. Stripe checkout + subscription gate

    Five free audits per repo, then a paywall comment on the next PR with a one-click Stripe checkout link. Subscription state syncs back to the installation via the Stripe webhook.