improve-ic-skill

Pass

Audited by Gen Agent Trust Hub on Nov 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill performs several local repository operations to facilitate maintenance and testing:
  • Executes npm run validate to check skill consistency and metadata (Steps 2, 5, 8).
  • Uses git log and grep to analyze skill history and upstream tracking status (Step 1).
  • Runs node scripts/evaluate-skills.js to execute regression tests and trigger evaluations (Step 7).
  • Invokes skill-validator to perform quality scoring of the content (Step 5).
  • These commands are standard development practices within the context of managing a software repository and are appropriately scoped to the local environment.
  • [INDIRECT_PROMPT_INJECTION]: The skill's workflow involves ingesting external data to identify required improvements, which presents a surface for indirect prompt injection:
  • Ingestion points: The agent is instructed to read problem statements from user input, GitHub issue bodies, and upstream repository diffs (Step 0, Step 6).
  • Boundary markers: There are no explicit delimiters or "ignore previous instructions" warnings mentioned for the interpolation of this data into the skill files.
  • Capability inventory: The agent has the capability to modify repository files in the skills/ and evaluations/ directories and execute local scripts for validation.
  • Sanitization: No explicit sanitization or filtering is described; however, the skill mandates a strict "understand the problem first" step that requires the agent to verify the specific issue before making changes, reducing the risk of accidental execution of embedded instructions.
  • Given the intended use case for repository maintenance by authorized contributors, this surface is considered low risk and functionally necessary.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 5, 2026, 12:25 AM