flow-healer-triage

name	flow-healer-triage
description	Run this skill when a Flow Healer run fails or behaves unexpectedly and the user wants a fast, deterministic diagnosis. Use for requests like "triage this healer failure", "why did the run fail", "classify this issue state", or "tell me if this is an environment problem or a product bug".

Flow Healer Triage

Use this skill after any failed or suspicious Flow Healer attempt.

Inputs

• --db-path
• --issue-id

Outputs

The script emits a JSON object with:

• issue
• latest_attempt
• diagnosis

Key Output Fields

• diagnosis
• latest_attempt.failure_class
• latest_attempt.failure_reason
• issue.state

Success Criteria

• Diagnosis is produced and paired with an operator-ready next action.

Failure Handling

• If the issue row is missing or incomplete, stop and repair state visibility before classifying further.
• Repair operator or environment problems before another live run.
• Treat connector_or_patch_generation as an immediate handoff to flow-healer-connector-debug.

Workflow

1. Run scripts/triage_issue.py with the healer DB path and issue id.
2. Read diagnosis first, then confirm it against latest_attempt.failure_class, latest_attempt.failure_reason, and issue.state.
3. If the failure lands in product_bug, read references/failure_catalog.md and compare against known incidents.
4. If the failure lands in operator_or_environment, fix the environment before another live run.

Default Command

.venv/bin/python skills/flow-healer-triage/scripts/triage_issue.py \
  --db-path /tmp/flow-healer-state/repos/live/state.db \
  --issue-id 3

Diagnosis Buckets

• operator_or_environment
• repo_fixture_or_setup
• connector_or_patch_generation
• product_bug
• external_service_or_github

Next Step

• operator_or_environment: repair the environment and rerun flow-healer-preflight.
• repo_fixture_or_setup: repair the repo/setup and rerun flow-healer-local-validation.
• connector_or_patch_generation: hand off to flow-healer-connector-debug and compare proposer versus verifier output contracts.
• product_bug: capture evidence and escalate.
• external_service_or_github: retry later with an operator note.

Load On Demand

• Use references/failure_catalog.md when the script classifies the issue as product_bug or when the pattern is unfamiliar.