Skip to content

Experts

Experts wrap the agent loop with intake → triage → fix → verify → publish.

When should I use which expert?

See the Early Adopter Use Cases guide — each scenario explains why, the command, and the benefit.

code-agent experts list

bug-fix — CI failure auto-remediation
When CI failed (compiler, tests, coverage, lint)
Input --log FILE or --run-id GITHUB_RUN_ID
Verify --verify-cmd (should match CI) 
code-agent experts run bug-fix \
  --log /tmp/ci.log \
  --verify-cmd "pytest -q" \
  --dry-run

code-agent experts run bug-fix \
  --log /tmp/ci.log \
  --verify-cmd "pytest -q" \
  --publish

Parsers: TypeScript/tsc, ESLint, mypy, Rust, Go, pytest, coverage, Python tracebacks, generic.

Flow:

  1. Parse log → signals
  2. RCA — correlate with git diff + history
  3. Build fix task (RCA in MR body)
  4. Agent loop + verify_cmd
  5. Optional publish → branch ci-fix/<run_id>-<fp>
Output Success Skipped Failed
status done skipped failed
signal_count ≥ 1 0 or duplicate ≥ 1
mr_url if --publish prior MR if duplicate —

Artifacts: signals.json, rca.json, fix_request.json, trace.json, diff.patch

code-review — PR inline line comments
When First-pass review on a PR diff (nits, bugs, security smells)
Input --pr N (required)
Output One GitHub Review with inline comments (side=RIGHT)
Publish Comment-only — does not open an MR 
code-agent experts run code-review --pr 42 -w .
code-agent experts run code-review --pr 42 --dry-run -w .   # print findings, no API post

Findings: .code-agent/runs/<run_id>/review-findings.json. Lines not in the PR diff are skipped. Invalid JSON → exit 2 (fail closed).

test-intel — intelligent test selection
When Speed up PR CI — run only impacted tests
Input Git diff vs main (no log required) 
code-agent experts run test-intel
code-agent experts run test-intel --base-branch main

Does not invoke the coding agent — produces a test plan only.

Output Description
plan.verify_cmd e.g. pytest -q tests/test_foo.py …
plan.impacted_tests Pytest node ids from diff
plan.shards Parallel shard groups
Artifact .code-agent/runs/<run_id>/test_plan.json

deploy-guard — continuous verification
When After deploy / canary — compare metrics to baseline
Input --metrics-file FILE 
code-agent experts run deploy-guard \
  --metrics-file metrics/current.json

Metrics JSON:

{
  "error_rate": 0.01,
  "latency_p99": 110.0,
  "saturation": 0.35
}
Decision Meaning
pass No anomalies
block Anomaly; no rollback (default)
rollback Only with auto_rollback + --confirm-rollback

sre-expert — incident response
When Alertmanager or generic webhook payload
Input --log ALERT.json 
code-agent experts run sre-expert \
  --log alert.json \
  --dry-run

Formats: Alertmanager { "alerts": [...] } or generic { "kind": "incident", ... }.

MR branch prefix: sre-fix/…

monitoring-expert — observability audit
When Weekly audit or pre-release instrumentation check
Input None (scans repo) 
code-agent experts run monitoring-expert --dry-run

Finds missing Prometheus/OTel metrics and invalid alert rules.

Shared expert flags

Flag Effect
--dry-run No publish; may still write locally
--publish Commit + draft MR
--no-artifacts Skip .code-agent/runs/
-w Workspace root

See Commands for the full flag list.