Totem Shield

Version updated for https://github.com/mmnto-ai/totem to version @mmnto/cli@1.19.0.

  • This action is used across all versions by 0 repositories.

Action Type

This is a Composite action.

Go to the GitHub Marketplace to find the latest changes.

Action Summary

Totem is a CLI tool designed to provide a persistent memory and enforcement layer for AI coding agents, addressing their tendency to forget architectural context and repeat mistakes. It automates the process of converting plain-English lessons into enforceable rules that catch and block repeated errors during development, ensuring architectural integrity and reducing friction in code reviews. Key capabilities include offline linting, deterministic rule enforcement, and a scalable mechanism to prevent recurring issues.

What’s Changed

Minor Changes

  • 9686817: ADR-091 Stage 4 Verify-Against-Codebase verifier (mmnto-ai/totem#1682). Headline 1.16.0 substrate. Before a compiled rule moves to Active status, it runs deterministically (zero LLM) against the consumer’s working tree and is routed into one of four outcomes:

    • No matches → status: 'untested-against-codebase'. Verifier ran, found no hits in this codebase. Subsequent compile cycles in a populated repo can re-run Stage 4 and promote.
    • Out-of-scope baseline match → status: 'archived', archivedReason cites the offending paths, new reasonCode 'stage4-out-of-scope-match' added to NonCompilableReasonCodeSchema. Pattern is over-broad.
    • In-scope badExample-shape match → new confidence: 'high' field set. Pattern fires on real code in the exact authored shape.
    • Candidate Debt (in-scope but not bad-example shape) → force severity: 'warning' so the rule is alive but cannot break CI on first run; totem doctor (T4 / mmnto-ai/totem#1685) will surface the candidate-debt sites.

    T1 ships local-compile mode fully. The verifier walks the consumer’s git-enumerated file set and reuses the existing applyRulesToAdditions / applyAstRulesToAdditions rule-engine surfaces; baseline glob matching mirrors the test-contract scope classifier shapes (mmnto-ai/totem#1626 / mmnto-ai/totem#1652). Telemetry events tagged type: 'stage4-verify' append to .totem/temp/telemetry.jsonl.

    Public API (new, exported from @mmnto/totem): verifyAgainstCodebase, getDefaultBaseline, DEFAULT_BASELINE_GLOBS, type Stage4VerificationResult, type Stage4Baseline, type Stage4Outcome, type Stage4VerifierDeps. New optional verifyStage4 callback on CompileLessonDeps; new optional onStage4Outcome callback on CompileLessonCallbacks. Schema deltas: CompiledRuleSchema.status enum gains 'untested-against-codebase', new optional confidence field, NonCompilableReasonCodeSchema gains 'stage4-out-of-scope-match'.

    Negative scope (deferred to follow-on tickets): Pack install→lint promotion + pending-verification state writers + .totem/rule-metrics.json reads belong to T3 (mmnto-ai/totem#1684). Consumer .totemignore / review.stage4Baseline config field belongs to T2 (mmnto-ai/totem#1683). totem doctor Stage 4 surfaces belong to T4 (mmnto-ai/totem#1685). Batched / streamed perf optimizations belong to T5 (mmnto-ai/totem#1686). Pipeline 1 (manual) rules bypass Stage 4 — those are human-authored and self-evidencing.

    Bot-review tail: Sonnet pre-push review caught a defensive-coding gap on the candidate-debt severity downgrade (treating undefined severity as a no-op would let pre-1.16.0 persisted rules skip the downgrade if a future lint pass interpreted undefined as 'error'). Tightened to severity !== 'warning' so the post-condition is explicit.

Patch Changes

  • Updated dependencies [9686817]
    • @mmnto/totem@1.19.0