skills/v1r3n/dinesh-gilfoyle/dg/Gen Agent Trust Hub

dg

Pass

Audited by Gen Agent Trust Hub on Oct 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local environment to be processed by agents.
  • Ingestion points: SKILL.md (Step 1: Gather Code) reads output from git diff and local files/directories.
  • Boundary markers: The skill provides a system-level instruction to sub-agents: "You are doing research only — read the code and produce your review. Do NOT edit any files." However, there is no explicit sanitization or delimiters for the code content itself.
  • Capability inventory: The skill possesses significant capabilities including shell command execution (git, gh, curl, npx) and file system access.
  • Sanitization: No sanitization, escaping, or filtering of the ingested code content is performed before passing it to the sub-agents.
  • [EXTERNAL_DOWNLOADS]: The skill downloads and executes remote content during the comic generation phase.
  • Finding: In SKILL.md (Step 4), the skill uses npx --yes puppeteer-screenshot to download and run a package from the npm registry. While npm is a well-known service, the use of npx --yes allows for the execution of remote scripts.
  • Finding: The skill fetches vulnerability data from api.osv.dev (Google) and services.nvd.nist.gov (NIST) using curl. These are well-known and trusted services for security metadata.
  • [COMMAND_EXECUTION]: The skill heavily relies on local shell command execution to perform its primary functions.
  • Finding: SKILL.md executes git diff, git diff --staged, and various ecosystem-specific audit tools (e.g., npm audit, pip audit, govulncheck).
  • Finding: The skill interacts with GitHub using gh pr view and gh pr comment to retrieve metadata and post review summaries/images.
  • [DYNAMIC_EXECUTION]: The skill generates and renders dynamic content at runtime.
  • Finding: SKILL.md populates comic-template.html with data extracted from the agent debate (banter, scores, targets) and saves it as a temporary file (/tmp/dg-comic-{timestamp}.html). This file is then rendered using a headless browser or system browser to create a screenshot.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 9, 2026, 04:02 PM