auto-skill-capture

Pass

Audited by Gen Agent Trust Hub on Aug 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill is designed to run a shell command specified in a local configuration file (SELF-IMPROVE.md) through the placeholder.\n
  • Evidence: SKILL.md directs the agent to "run the resolved " following changes to global skills. The example in references/self-improve-example.md demonstrates using npx to execute a CLI tool.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes existing skill files to manage updates and curation, which exposes the agent to instructions embedded in those files.\n
  • Ingestion points: Reads SKILL.md and related files from the paths resolved for and .\n
  • Boundary markers: None. The instructions do not mandate the use of delimiters or provide warnings to ignore embedded instructions when reading existing skills.\n
  • Capability inventory: Includes file system read/write and shell command execution.\n
  • Sanitization: None. There is no mention of sanitizing or validating the content extracted from existing skill files.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 28, 2026, 01:03 PM
Security Audit — agent-trust-hub — auto-skill-capture