skills/rpamis/comet/comet-build/Gen Agent Trust Hub

comet-build

Warn

Audited by Gen Agent Trust Hub on May 29, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes multiple shell commands to manage environment state, git branches, and project-specific build operations.\n- [REMOTE_CODE_EXECUTION]: The skill dynamically locates and sources an environment script (comet-env.sh) by searching through the current directory and several hidden folders in the user's home directory (e.g., $HOME/.*/skills, $HOME/.config, $HOME/.gemini). Sourcing and executing scripts from paths discovered at runtime is a dangerous pattern that can lead to arbitrary code execution if malicious files are present in the search paths.\n- [PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface by ingesting and following instructions from external plan files and task lists.\n
  • Ingestion points: Reads feature plans from docs/superpowers/plans/ and task lists from tasks.md.\n
  • Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands within the ingested data.\n
  • Capability inventory: Full shell access, git repository management, and the ability to load and execute other agent skills.\n
  • Sanitization: No evidence of validation or sanitization of the content within the plan or task files before processing.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 29, 2026, 11:22 AM
Security Audit — agent-trust-hub — comet-build