search-cold-skills

Warn

Audited by Gen Agent Trust Hub on Jul 8, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill requires the agent to run an installation script (scripts/install.sh) and a search helper (scripts/query.sh). These scripts perform file system modifications (mkdir, sed, rm) and change file permissions (chmod +x) within the agent's environment.
  • [PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by instructing the agent to 'read that cold skill's SKILL.md fully' and 'Follow the cold skill's instructions manually.' Malicious content within a 'cold skill' could override agent behavior or execute unintended tasks once loaded into the active context.
  • [REMOTE_CODE_EXECUTION]: The skill employs dynamic execution patterns by resolving and executing script paths (scripts/, references/) defined within external cold skill directories at runtime.
  • [COMMAND_EXECUTION]: The scripts/query.sh script processes user-supplied queries through shell expansion in loops (for w in $QUERY) and regex tools (rg), which could lead to unexpected behavior if the query contains shell metacharacters.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Jul 8, 2026, 09:16 AM
Security Audit — agent-trust-hub — search-cold-skills