thanhnvandClaude Opus 4.8
1b61d7f381
Add a shared log taxonomy (error<warn<info<debug<trace, default info) wired through all three layers, plus a machine-readable per-step run log. - scripts/casan-log.mjs + .specify/scripts/bash/casan-log.sh: shared logger ([LEVEL] ts [component] msg), stderr-only so stdout contracts and exit codes are byte-identical. Node side includes trace-level redaction (secret/PII). - run-casan-pipeline.mjs: per-STEP info line, loop-activation warnings (BACK-TO-PLAN, FAIL->STEP6), debug harness rc + trace_id, trace payload excerpt, end-of-run 13-STEP summary table, and one JSONL line/step in .specify/logs/pipeline-run.jsonl. New --dry-run stubs all agents but keeps the real wrapper in the loop (deterministic, offline, full 13 STEP + loops). - casan-step.mjs: debug logs for judge verdict, checkpoint, rollback; logger import degrades to noop when the file is copied standalone (T1 test). - casan-harness.sh: debug-log each phase H4-in -> H5 -> [H2-gate] -> H6-exec -> H4-out with its rc; optional CASAN_PHASE_REPORT JSON for the Boss. Default level (info) keeps output close to before; behavior opt-in via env. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>