solution-knowledge-base

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local shell commands such as git diff and git log to extract context, and calls a local validation script python scripts/validate_solution_frontmatter.py to check file integrity.
  • [PROMPT_INJECTION]: The skill modifies the repository's agent instruction files (AGENTS.md and CLAUDE.md) with instructions that direct the agent to consult the solution store during future planning or review sessions.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by ingesting data from untrusted sources.
  • Ingestion points: Extracts problem details and resolutions from the current conversation context, git diff, and git log (SKILL.md, Step 1).
  • Boundary markers: Uses explicit marker blocks (<!-- NEXUS_SOLUTIONS_START -->) when modifying instruction files (SKILL.md, Step 5), but no specific markers are mentioned for the content extracted from logs and written to documentation.
  • Capability inventory: Writes files to the local filesystem (docs/solutions/), modifies instruction files, and executes a Python script (SKILL.md, Steps 4, 5, 6).
  • Sanitization: There is no description of content sanitization or escaping for data gathered from git logs or the session history before it is persisted.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 10:15 AM
Security Audit — agent-trust-hub — solution-knowledge-base