fix(harness/opencode): surface stderr Error: lines on exit 0

[AbirAbbas]

Summary

The opencode CLI sometimes prints a hard error to stderr but exits 0 — notably:

• Error: Model not found: … (invalid model id)
• AuthenticationError (bad/missing key)
• APIError (provider failure)

OpenCodeProvider._execute_impl only flagged failures on non-zero exit, so these cases returned RawResult(is_error=False, result=None) — silently empty output. Downstream callers (the schema-output retry loop in agentfield.harness._runner) then surface the failure as the much less useful "agent failed to produce a valid result".

The bug is doubly hidden because clean_stderr is logged truncated to 800 chars, but opencode opens stderr with the SQLite migration prelude (Performing one time database migration…), pushing the real Error: line off the end of the truncation window. Operators saw the prelude in logs and assumed migration was the issue, not the model id.

This was the proximate cause of the SWE-AF Railway run run_1777925624226_f5a637ae failing every run_product_manager and run_git_init invocation in 8–10s with no useful diagnostic. A separate companion PR (Agent-Field/SWE-AF#55) fixes the upstream model-string bug; this PR ensures opencode failures of this shape are never invisible again.

Changes

• sdk/python/agentfield/harness/providers/opencode.py

  • New _OPENCODE_STDERR_ERROR_PATTERNS and _extract_opencode_error() helper that pulls the meaningful failure line + a small context window out of opencode stderr, skipping the migration prelude.
  • Adds an explicit branch for returncode == 0 and result_text is None and stderr matches an error pattern → returns is_error=True, failure_type=CRASH with the extracted error message.
  • The existing non-zero-exit branch now uses _extract_opencode_error() too, so log truncation can no longer hide the cause there either.

• sdk/python/tests/test_harness_provider_opencode.py — three new regression tests:

  1. Model not found exit 0 → is_error=True with the model id in the message.
  2. Migration-only stderr + real stdout → is_error=False (no false positive).
  3. Long migration prelude + AuthenticationError on non-zero exit → AuthenticationError surfaces (no truncation).

Test plan

• pytest tests/test_harness_provider_opencode.py tests/test_harness_factory.py tests/test_harness_runner.py — 48/48 pass.
• Bump SWE-AF's agentfield requirement to whatever version this lands in, redeploy, re-trigger github buddy implement on Agent-Field/github-buddy#22, confirm any future opencode misconfiguration shows up with the actual error in the Railway logs (not buried under the migration prelude).

Companion PR

Agent-Field/SWE-AF#55 — fixes the upstream cause (default model id was missing the openrouter/ provider prefix). Both PRs are needed to fully unblock the github-buddy → SWE-AF delegation flow on Railway.

🤖 Generated with Claude Code

[github-actions]

Performance

SDK	Memory	Δ	Latency	Δ	Tests	Status
Python	7.9 KB	-13%	0.50 µs	+43%	✓	⚠

✓ No regressions detected

[github-actions]

📊 Coverage gate

Thresholds from .coverage-gate.toml: per-surface ≥ 86%, aggregate ≥ 88%, max per-surface regression ≤ 1.0 pp, max aggregate regression ≤ 0.50 pp.

Surface	Current	Baseline	Δ
control-plane	87.40%	87.30%	↑ +0.10 pp	🟡
sdk-go	91.80%	90.70%	↑ +1.10 pp	🟢
sdk-python	93.66%	93.63%	↑ +0.03 pp	🟢
sdk-typescript	92.63%	92.56%	↑ +0.07 pp	🟢
web-ui	89.69%	90.01%	↓ -0.32 pp	🟡
aggregate	88.85%	89.01%	↓ -0.16 pp	🟡

✅ Gate passed

No surface regressed past the allowed threshold and the aggregate stayed above the floor.

[github-actions]

📐 Patch coverage gate

Threshold: 80% on lines this PR touches vs origin/main (from .coverage-gate.toml:thresholds.min_patch).

Surface	Touched lines	Patch coverage	Status
control-plane	0	—	➖ no changes
sdk-go	0	—	➖ no changes
sdk-python	0	—	➖ no changes
sdk-typescript	0	—	➖ no changes
web-ui	0	—	➖ no changes

✅ Patch gate passed

Every surface whose lines were touched by this PR has patch coverage at or above the threshold.