validate-criterion

After executing work, call this to validate each acceptance criterion. Statuses passed and failed require concrete evidence from actual execution.

💡 Fast-path: Use the run_and_validate tool to automatically execute a shell command and record stdout/stderr/exitCode as high-confidence evidence in a single call.

Evidence Requirements

You MUST run the actual command or check before validating. Do NOT validate based on memory, assumption, or "it should work".

method	What to do	Evidence to provide
command	Run the exact command in a shell	Paste the real stdout/stderr + exit code
file	Read the file and check the content	Paste the relevant lines from the file
url	Make the HTTP request	Paste the response status + body

What NOT to do

• ❌ Validate passed without running anything
• ❌ Write "should work" or "looks correct" as evidence
• ❌ Copy evidence from a previous run without re-running
• ❌ Use evidence_type=text for a criterion with method=command
• ❌ Validate before implementation is actually complete

Self-Claimed Status

When role=agent (default), passed criteria are marked selfClaimed=true. This means can_complete_goal will block until an independent reviewer calls confirm_criterion with fresh evidence.

This is by design — your self-assessment is not trusted for completion. Use confirm_criterion (as a separate reviewer agent) to convert self-claimed passes to formal passes.

Example

{
  "criterion_id": "api-200",
  "status": "passed",
  "evidence": "curl -s localhost:3000/health → 200 {\"status\":\"ok\"}",
  "evidence_type": "command"
}

{
  "criterion_id": "api-500",
  "status": "failed",
  "evidence": "curl -s localhost:3000/users → 500 Internal Server Error: TypeError: Cannot read property 'name' of undefined",
  "evidence_type": "command"
}