skills/dcoon/agent-kit/agent-skill/Gen Agent Trust Hub

agent-skill

Pass

Audited by Gen Agent Trust Hub on Nov 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses several CLI tools to perform its core functions.
  • Executes npx skills to manage skill installations and synchronizations (references/npx-skills.md).
  • Executes uvx --offline lintlang to perform syntax and best-practice checks on skill files (references/lint.md).
  • Executes git pull to synchronize local repositories (references/compress.md).
  • Executes wc -lc to provide token count and file size comparisons (references/refactor.md).
  • [REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of external code.
  • Uses npx to run the skills package from the npm registry.
  • Uses uvx to download and run the lintlang package from the Python Package Index (PyPI).
  • The linter tool is sourced from github.com/hermes-labs-ai/lintlang.
  • [EXTERNAL_DOWNLOADS]: The skill fetches documentation and configuration from external sources.
  • References documentation at github.com/vercel-labs/skills (Trusted Organization).
  • References documentation and specifications at opencode.ai and agentskills.io.
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to process, lint, and refactor external SKILL.md files, creating a surface for indirect prompt injection.
  • Ingestion points: Reads SKILL.md and related reference files from the filesystem based on user goals (SKILL.md, references/lint.md).
  • Boundary markers: Uses pseudo-Makefile structures with explicit 'safety sentinels' (@echo ...; exit 1) and instructions to the agent to skip these lines to prevent accidental shell piping.
  • Capability inventory: Includes file system read/write access, network access via package managers (npx, uvx), and shell command execution (git, wc).
  • Sanitization: Implements a dedicated linter (lintlang) to validate formatting and check for redundant or non-imperative phrasing, and requires explicit user consent before applying any file edits.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 21, 2026, 04:19 PM