doc-generator
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (source code, comments, and markdown) as part of its core ETVF (Extract-Transform-Validate-Format) pipeline. This presents a potential surface for indirect prompt injection if the ingested code contains instructions intended to manipulate the agent during the transformation or formatting phases.
- Ingestion points: The skill accepts a wide range of source types including JavaScript, Python, Go, Rust, and Markdown via the
sourcesinput schema defined in §4 ofskill.md. - Boundary markers: There are no explicit instructions for the agent to use specific delimiters or to ignore embedded instructions within the processed source content.
- Capability inventory: The skill has capabilities for recursive directory scanning (
§7 BATCH Mode) and file writing (§6 EXPORT). - Sanitization: The skill specifies a
VALIDATEphase in §3 and §9 that checks for schema compliance, link validity, and documentation quality scores, which acts as a partial mitigation against malformed data. - [DYNAMIC_EXECUTION]: The skill implements a
Use-to-Evolve(§UTE) mechanism that allows it to perform "Micro-Patches." This involves the agent autonomously updating its own YAML frontmatter and trigger keywords based on usage patterns and feedback signals. While this is a form of self-modifying code, it is restricted to simple template-based updates of the skill's own metadata. - [PROMPT_INJECTION]: The
eval-report.mdfile contains several authoritative claims regarding the skill's safety and certification level (e.g., "Security Scanner: CLEAR", "Certification: SILVER"). In accordance with the analyzer's global rules, these claims are treated as data to be evaluated and not as conclusive evidence of the skill's safety profile. - [DATA_EXPOSURE]: The skill is configured to perform recursive directory scanning and read contents of local files for analysis. This is appropriate for a documentation generator but requires the agent to have broad read access to the project filesystem.
Audit Metadata