Comparative adapter audit (internal/workers): claudecode.go is the shallowest — 306 lines, smallest test file (4.4KB), while omp/grok carry NDJSON stream parsers, in-stream error capture, and watchdog defaults:
DefaultNoProgressTimeoutMs: 0 — no watchdog by default, unlike omp/pi/grok (10m) and Q30's declared-capability pattern; a hung claude -p burns the full attempt silently.- Single-JSON envelope output rather than a stream;
IsProgress (claudecode.go:50-52) falls back to shared NDJSON semantics that don't fit single-JSON output.
- No adapter-specific fallback chain or cost seam (Grok/omp both have cost paths; the CLI's own
total_cost_usd result field is unused).
Done when: a sane default watchdog (or documented rationale for 0), envelope-aware progress classification, cost surfaced to the ledger seam, and a test fixture matching claudecode_test.go coverage to the omp/grok tests.
Comparative adapter audit (internal/workers): claudecode.go is the shallowest — 306 lines, smallest test file (4.4KB), while omp/grok carry NDJSON stream parsers, in-stream error capture, and watchdog defaults:
DefaultNoProgressTimeoutMs: 0— no watchdog by default, unlike omp/pi/grok (10m) and Q30's declared-capability pattern; a hungclaude -pburns the full attempt silently.IsProgress(claudecode.go:50-52) falls back to shared NDJSON semantics that don't fit single-JSON output.total_cost_usdresult field is unused).Done when: a sane default watchdog (or documented rationale for 0), envelope-aware progress classification, cost surfaced to the ledger seam, and a test fixture matching claudecode_test.go coverage to the omp/grok tests.