spec
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill performs legitimate project documentation and specification management tasks. It uses standard developer tools (e.g., git, grep, rg) to analyze the local repository and generate documentation in markdown format. No external network connections or unauthorized access to sensitive files outside the project scope were detected.
- [DYNAMIC_EXECUTION]: The skill invokes local utility scripts and shell commands to facilitate its analysis and documentation workflows. These executions are based on project-relative paths and do not involve untrusted remote content.
- Evidence: Calls to local scripts such as
python .claude/scripts/code_graphfor dependency tracing andnode .claude/scripts/ai-signal-scan.cjsfor scanning AI-related patterns, as documented inreferences/author.mdandreferences/mode-discovery.md. - [INDIRECT_PROMPT_INJECTION]: The skill exhibits an inherent attack surface for indirect prompt injection because it ingests and processes project data (source code and existing specs) to generate specifications and test cases.
- Ingestion points: Codebase analysis, entity extraction, and operation discovery phases (Step 1-INIT.1, Phase A, B, and C) in
references/author.mdand spec-corpus scanning (Step 1) inreferences/mode-discovery.md. - Boundary markers: The skill enforces the use of structured 8-section templates and "tech-free" prose rules to decouple business intent from potentially malicious implementation details.
- Capability inventory: The skill can read local files, execute local analysis scripts (Code Graph, AI Signal Scan), and write markdown documentation to the project's documentation directories.
- Sanitization: It implements strict AI-SDD Mandates (M1-M7) and an "AI-Implementability Gate" to ensure that all authored content is re-implementable and verified against evidence, reducing the risk of accidental instruction obedience.
Audit Metadata