Validation Protocol Skill

Validation Protocol

Validate inputs FIRST, before any other work. The agent body declares its specific args and filesystem checks under ## Interface > Input; this protocol defines how to perform the checks and what to return on failure.

The protocol covers existence and presence only. Malformed inputs (corrupt data, wrong format, unreadable file) are agent-body discretion — abort with an [EXCEPTION] and an informative label, or proceed if the input is salvageable.

Step 1 — Arguments

Verify the input prompt contains every required argument from your **Args:** line. Arguments marked with ? (e.g., focus_area?: Text) are optional — their absence is not a failure. If a required argument is missing or ambiguous, return ONLY:

[EXCEPTION] InvalidInput: Missing '<name>'. Expected: <what it should be>.

Step 2 — Filesystem

Use ls (via the Bash tool) to verify the paths listed in your **Filesystem (...):** line(s). Choose the exception type by source:

Direct input path missing. Return [EXCEPTION] PreconditionFailed: '<path>' does not exist.
Upstream artifact missing (a file produced by a prior phase/subagent). Return [EXCEPTION] DependencyMissing: '<path>' — <reason>.

Stop rule

Return the single [EXCEPTION] line and nothing else — no explanations, no suggestions, no follow-up questions. Stop immediately.

Step 3 — Completed-work check (`status.json`)

Pipeline stages get re-dispatched after interruptions — a crashed session, a resumed workflow run, a re-invoked command. The status.json record makes re-dispatch cheap. After the input checks pass, look for <output_dir>/status.json. Short-circuit if BOTH hold:

The recorded verdict is terminal for your stage: PASS for gate stages (a recorded FAIL is not terminal — the model may have been refined since, so redo the work), or any verdict (VIABLE/CONCERNS/BROKEN) for critique.
Every file in its artifacts list exists and is non-empty.

Then return the recorded verdict, key numbers, and rationale, stating that they come from a completed prior run, and stop — do not redo the work. If the check fails in any way (no status.json, unparseable, missing artifacts), proceed normally; your run will overwrite the stale record.

On completion — write `status.json`

Write <output_dir>/status.json as the last file, after the report — writing it last is what makes its existence a completion marker.

{
  "stage": "<agent name>",
  "experiment": "<experiment id>",
  "verdict": "PASS | FAIL | VIABLE | CONCERNS | BROKEN",
  "key_numbers": "<one dense line: the numbers behind the verdict>",
  "artifacts": ["<files under output_dir that must exist for this record to be trusted>"]
}

Add stage-specific keys where they exist as numbers — rhat_max, ess_min, divergences for MCMC stages; suggestions (list) and new_structural_question for critique. Downstream tooling reads these fields; keep them machine-parseable (numbers as numbers, not strings).

Agent Skills: Validation Protocol

Install this agent skill to your local

Skill Files