Skip to content

Loki Mode v7.117.0

Choose a tag to compare

@github-actions github-actions released this 02 Jul 16:46
Immutable release. Only release title and notes can be modified.

Trust + adoption batch (zero-test hardening + rank 13; RED/GREEN, council-gated)

  • Anti-fake-green: a runner that executed ZERO tests now records INCONCLUSIVE,
    not pass.
    Completing the v7.116 node --test arc: a *.test.js with no test()
    calls (node --test exits 0, "# tests 0") and jest --passWithNoTests with no
    suites previously recorded pass:true / passed_count:0 -- a mini fake-green.
    Fixed across ALL runners in BOTH the recording paths (autonomy/run.sh in-loop
    test gate + autonomy/verify.sh) AND the READING path (autonomy/completion-council.sh
    evidence gate): a real runner that ran zero tests records
    {pass:"inconclusive", status:"no_tests_run"}, and the completion gate routes
    that to INCONCLUSIVE (pass-through, NOT affirmative green, NOT a block) rather
    than counting the string as pass. A real test file with >=1 passing test still
    records pass (unchanged); a failing suite still blocks (unchanged). This is a
    shipping-contract change (a repo with jest --passWithNoTests and no tests goes
    from VERIFIED to inconclusive -- intended, honest, source_without_runnable_tests).
    Tests: tests/test-zero-test-inconclusive.sh (13/13) + Case 8 in
    test-evidence-gate-no-tests.sh (29/29; a targeted revert of only the gate proves
    the gate fix is load-bearing).
  • rank 13 -- loki heal --assess read-only modernization readiness triage. A
    new read-only subcommand that scans a codebase and emits a readiness report:
    LOC / language mix, technical-debt signals, maintenance-risk indicators, a
    modernization maturity-level placement, and ranked candidate targets with
    rationale -- all from real deterministic scans (never fabricated numbers).
    Proven READ-ONLY: git status stays clean and no .loki dir is created in the
    target after --assess. Test: tests/test-heal-assess-readiness.sh (9/9, incl. the
    read-only assertion).

Both tests wired into run-all-tests.sh + local-ci.sh. The zero-test verdict-path
change was confirmed by a live calibration of the shipped artifact: a real repo
with a *.test.js containing zero test() calls recorded runner:node-test,
status:inconclusive ("runner ran but executed zero tests"), verdict CONCERNS -- no
fake-green. Reviewed by a 3-reviewer council; a third item (grill-report wiring)
was deferred because it widened the surface of a separate, pre-existing no-HITL
hang on spec-internal contradictions.