optimize-claude-md

Pass

Audited by Gen Agent Trust Hub on May 17, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [DYNAMIC_CONTEXT_INJECTION]: The skill uses the !pwd command in SKILL.md to retrieve the working directory when the skill is loaded. This is a benign use of dynamic context injection intended to support path resolution for file operations.\n- [EXTERNAL_DOWNLOADS]: The skill invokes uvx skilllint@latest, which downloads and executes the latest version of the skilllint package from the Python Package Index (PyPI). This introduces a runtime dependency on an external, unversioned package.\n- [COMMAND_EXECUTION]: The agent is instructed to perform shell command execution via uvx to audit skill completeness and measure token counts during the optimization process.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) because it reads and processes external documentation content.\n
  • Ingestion points: Phase 3 involves reading the full content of user-specified files (e.g., CLAUDE.md, SKILL.md) which may originate from untrusted sources.\n
  • Boundary markers: The skill does not provide instructions to the sub-agents to treat the ingested documentation as untrusted data or to ignore directives embedded within that data.\n
  • Capability inventory: The skill has the capability to spawn sub-agents (delegation), execute shell commands (via uvx), and write files to the filesystem upon user approval.\n
  • Sanitization: No sanitization or validation of the documentation content is performed before it is passed to the sub-agents for analysis and transformation.
Audit Metadata
Risk Level
SAFE
Analyzed
May 17, 2026, 04:04 AM
Security Audit — agent-trust-hub — optimize-claude-md