mm_verify_parse_adversarial

Parse one adversarial-verifier agent output blob into a tagged outcome (finding | null_result | malformed). Mirrors Rust parse_adversarial_output byte-for-byte. Used by /mm:verify-spec —iterative to sweep the 3-prompt adversarial bank per run; the slash command routes findings into classify-findings, records null_results as the convergence signal, and halts on malformed. The parser forces id="" and tier=null on any returned Finding (classifier owns both).

At a Glance

• Category: Plugin Management
• Source: Built-in
• Required parameters: 2
• Total parameters: 2

Parameters

Parameter	Type	Required	Description
raw	string	Yes	The raw agent output to parse.
prompt_id	string (enum: violation_test, production_input, breaking_assumption)	Yes	Which of the 3 adversarial bank prompts this output came from. Used to validate the null token matches (e.g. NO_VIOLATION_FOUND for violation_test).

Example

{
  "tool": "mm_verify_parse_adversarial",
  "arguments": {
    "raw": "example-raw",
    "prompt_id": "violation_test"
  }
}

Related Tools

• mm_list_plugins - List all loaded plugins with their id, name, version, category, tier, and too…
• mm_get_plugin_info - Get detailed information about a specific plugin including its tools, version…
• mm_push_team_checkin - Push a team check-in to the cloud. Scans local specs and features, captures c…
• mm_fetch_team_checkins - Fetch all team members’ check-ins from the cloud. Returns display names, bran…
• mm_claim_spec - Claim a spec from a teammate. Records the claim in the cloud (first-claim-win…