AI Agent Governance Framework enforcement interface. Exposes SKILL constraints as MCP resources and tools for audit, review, and security enforcement.
Two tools with partial schemas and descriptions. submit_review has a clear description but minimal parameter documentation. execute_tool has vague descriptions and underdocumented parameters. Neither tool has documented output schemas. Parameter descriptions lack format constraints, ranges, or validation guidance. Error handling is present but generic. No tool annotations (readOnlyHint/destructiveHint). Naming is reasonable (verb_noun) but descriptions fall short of LLM-optimization standards (10-1024 chars; these are ~150-200 chars, acceptable but sparse on actionable detail).
Request OS action execution. Internally checked by SecurityGuard. Checks command and path against security rules from security.yaml.
Submit content for audit. Internal check chain (invisible to Agent): 1. Load guidelines.yaml (invisible to AI) 2. Load template.yaml (for format comparison) 3. Reviewer.audit(content) — 5 static checks 4. All pass → APPROVED / Violations → P8AuditError
execute_tool parameters lack actionable descriptions. 'skill' defaults to 'prd' but no explanation of what skills are or how to discover them. 'operation' enum (read, write, delete, create) is not documented in the parameter description itself, LLM must infer from the name.
No output schemas documented for either tool. submit_review returns a dict with 'error', 'passed', 'score', 'violations', 'action', but this structure is not declared in the tool definition. execute_tool returns 'command_check', 'path_check', 'allowed', 'action', also undocumented. LLMs cannot plan downstream calls without knowing response structure.
Missing tool annotations. execute_tool is marked Risk: WRITE but has no destructiveHint annotation in the schema. submit_review is marked Risk: READ_ONLY but has no readOnlyHint. These hints are required for agents to reason about side effects and retry safety.
Inferred effective spec: <=2025-11-25.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 49 | <=2025-11-25 | v2 |
Parameter 'skill' in both tools defaults to 'prd' with no explanation of what 'prd' means, how to discover available skills, or what happens if an invalid skill is passed. The resource skill://index exists but is not mentioned in tool descriptions.
Error handling is generic. P8AuditError returns violations array but no guidance on how to fix them or what each violation means. execute_tool returns 'BLOCKED, You MUST abort' but does not explain why the command/path was blocked or what alternatives exist.