optimize-claude-md

Pass

Audited by Gen Agent Trust Hub on Mar 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Executes uvx skilllint@latest to perform file checks, measure token counts, and audit skill completeness.
  • [COMMAND_EXECUTION]: Employs dynamic context injection (!pwd) in SKILL.md to resolve the current working directory during skill initialization.
  • [EXTERNAL_DOWNLOADS]: Fetches the skilllint package from a standard registry using uvx to ensure the latest version is used for analysis.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface (Category 8) because it reads and processes user-provided files that are then analyzed by an LLM agent. (1) Ingestion points: Target files and directories provided via $ARGUMENTS are processed in Phase 3. (2) Boundary markers: The delegation template uses structural separators but lacks specific instructions to ignore embedded commands within the target files. (3) Capability inventory: Performs command execution via uvx in SKILL.md. (4) Sanitization: Includes an independent verification step by a second agent (Phase 5) and requires explicit human approval (Phase 8) before any modifications are written to disk.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 29, 2026, 08:41 AM