Runtime Sweep

Runtime Sweep

Announce at start: "I'm using the runtime-sweep skill to verify runtime registration and wiring integrity."

Usage

/runtime-sweep
/runtime-sweep --dry-run
/runtime-sweep --scope omnidash-only

Execution

Phase 1 — Parse arguments

• --dry-run → pass through to node
• --scope → all-repos (default) or omnidash-only

Phase 2 — Run node

onex node node_runtime_sweep -- \
  --scope <all-repos|omnidash-only> \
  [--dry-run]

Capture stdout (JSON: RuntimeSweepResult). Exit 0 = clean, exit 1 = findings.

On non-zero exit, a SkillRoutingError JSON envelope is returned — surface it directly, do not produce prose.

Phase 3 — Render report

From the JSON output display four summary tables:

Node Descriptions — REAL / PLACEHOLDER / MISSING per contract

Handler Wiring — WIRED / UNWIRED / ORPHAN_TOPIC per handler file

Topic Symmetry — SYMMETRIC / PRODUCER_ONLY / CONSUMER_ONLY per topic

Report counts by finding type. List each finding with subject and type.

Phase 3b — Docker Log Analysis

For each running container, call get_container_logs() via docker_helper:

logs = docker_helper.get_container_logs(container_id, tail=500)

Classify container log health:

• CLEAN: no errors in last 500 lines
• NOISY: repeated non-fatal warnings
• ERROR_HEAVY: >10% of lines are errors
• CRASH_LOOP: container restarted in last 5 minutes

Phase 3c — Live Container Crash-Loop Detection

For each omninode-runtime* container, run:

docker inspect --format='{{.RestartCount}}' "$CONTAINER_NAME"

Threshold: RestartCount > 5 → CRASH_LOOP_CRITICAL finding. Writes friction YAML to $ONEX_STATE_DIR/friction/runtime_sweep_<timestamp>.yaml and exits non-zero.

Phase 3d — Orchestrator Consumer Group Health

Derive consumer groups from event_bus.subscribe_topics in contract YAMLs. For each group:

docker exec omnibase-infra-redpanda rpk group describe "$GROUP_NAME" --brokers localhost:9092

STATE: Empty → CONSUMER_GROUP_EMPTY finding. Writes friction YAML and exits non-zero.

Phase 4 — Ticket creation (skipped if --dry-run)

For each finding with type PLACEHOLDER_DESCRIPTION, MISSING_DESCRIPTION, UNWIRED_HANDLER, ORPHAN_TOPIC, PRODUCER_ONLY, CONSUMER_ONLY, CRASH_LOOP_CRITICAL, or CONSUMER_GROUP_EMPTY, create a Linear ticket via tracker.save_issue:

Title: fix(wiring): <finding_type> — <subject>
Project: Active Sprint
Labels: wiring, runtime-sweep

Skip ticket creation for REAL, WIRED, SYMMETRIC findings (healthy state).

Phase 5 — Write skill result

Write to $ONEX_STATE_DIR/skill-results/<run_id>/runtime-sweep.json:

{
  "skill": "runtime-sweep",
  "status": "clean | findings | error",
  "contracts_checked": 0,
  "total_findings": 0,
  "by_type": {}
}

If Phase 3c or Phase 3d found violations, additionally write a friction YAML to $ONEX_STATE_DIR/friction/runtime_sweep_<timestamp>.yaml with crash-loop and consumer-group findings, then exit non-zero.

Architecture

SKILL.md  → thin shell: parse args → node dispatch → render results
node      → omnimarket/src/omnimarket/nodes/node_runtime_sweep/
contract  → node_runtime_sweep/contract.yaml

Routing contract: dispatch must use onex node <node_name> (not onex run). Non-zero exit emits a SkillRoutingError JSON envelope — callers must surface it verbatim, never paraphrase.

All wiring verification logic lives in the node handler. This skill does no analysis.