skills/cloudvoyant/codevoyant/task/Gen Agent Trust Hub

task

Pass

Audited by Gen Agent Trust Hub on Aug 21, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill reads and acts upon task definitions found in local project files such as 'package.json' or 'mise.toml'. A maliciously crafted repository could contain task names or definitions designed to execute unauthorized commands if the user or agent triggers them.
  • Ingestion points: The script 'scripts/task.sh' reads task names from 'package.json', 'mise.toml', 'justfile', and 'Taskfile.yml'.
  • Boundary markers: None present; the skill treats task names from these files as literal strings to be executed.
  • Capability inventory: The skill is capable of executing shell commands through 'exec' using system task runners like 'npm', 'pnpm', 'mise', 'just', or 'task'.
  • Sanitization: The script uses literal matching and shell-safe quoting for arguments, though the final command interpretation is handled by the underlying runner specified in the project configuration.
  • [PERSISTENCE_MECHANISMS]: The skill includes logic to modify project configuration files (e.g., adding a task to 'mise.toml' or 'package.json'). This is an intended feature that allows the agent to help set up a project's workflow, and it includes a requirement for explicit user confirmation before any file writes occur.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 21, 2026, 02:36 PM
Security Audit — agent-trust-hub — task