Skip to content

feat: offline evidence CLI for agents and CI (v1.23.0) - #54

Merged
alloevil merged 1 commit into
masterfrom
feat/offline-evidence-cli
Sep 24, 2026
Merged

alloevil merged 1 commit into
masterfrom
feat/offline-evidence-cli

Conversation

@alloevil

Copy link
Copy Markdown
Owner

Direction

Make the existing evidence rules usable by agents/local automation without a browser, a server or a model judge. No new rule implementation or runtime dependency.

Changes

  • Add inspect --platform omp|codex|claude-code FILE with minimized versioned JSON or text.
  • Read one stable regular UTF-8 JSONL input, capped at 64 MiB; fail explicitly on invalid/changed input or known normalization loss.
  • Keep raw logs/arguments/paths/IDs/notes out of reports; include input/rule/adapter hashes and original physical source lines.
  • Exit 0 means complete report, never task success; optional --fail-on pending-failures uses exit 2; input/coverage errors use 1.
  • Generate and package CommonJS rules from UI TypeScript; add a CI drift check. Existing UI rules and web behavior unchanged.

Verification

  • 332 tests pass, including 19 new CLI cases: privacy, no-network/server imports, read-only HOME/input, deterministic output, error policy, mixed/multi-result coverage gaps and UI parity.
  • 15 frozen real sessions / 2024 tool results match UI summaries; 2493 references verified; no raw evaluation data included.
  • Actual tarball inspect works without node_modules or frontend source/compiler.
  • UI build, lint (existing findings unchanged), actionlint, 16 automated claims and documentation targets pass.
  • Staged source secret scan passes; private output is not committed.

Scope

No databases/compressed files/recursive session discovery, no automatic remediation or coverage verdict. Reports are minimized, not guaranteed anonymous. Follow the existing protected release workflow after CI.

@alloevil
alloevil merged commit 45e9cb8 into master Sep 24, 2026
5 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant