skill-publish

Pass

Audited by Gen Agent Trust Hub on Sep 5, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external content (SKILL.md files from paths, URLs, or pastes) which could contain malicious instructions designed to influence the publishing process.\n
  • Ingestion points: Artifact input in Step 1 (folder path, direct file path, pasted content, or GitHub URL).\n
  • Boundary markers: Implements a mandatory Step 0 'QA gate' using /skill-qa and a backend CI check (validate-skill.ts) to validate content before publishing.\n
  • Capability inventory: Performs shell commands using gh (GitHub CLI) for API dispatches and issue comments, and uses tar/base64 for file processing (Step 5, Step 6).\n
  • Sanitization: Frontmatter parsing and sanity checks are performed before dispatching data to the remote repository.\n- [COMMAND_EXECUTION]: Uses various shell utilities including gh, npx, tar, and base64 to automate the GitHub workflow, package artifacts, and interact with repository issues as part of the intended publishing lifecycle.\n- [DATA_EXFILTRATION]: Reads local skill files and transmits them as Base64-encoded payloads to the author's specific GitHub repository (peter-tu-zynkr/zynkr-skill-builder) via API dispatches to initiate a Pull Request. This behavior is documented as the skill's primary function.\n- [EXTERNAL_DOWNLOADS]: Fetches skill content from user-provided GitHub URLs and utilizes tools from the author's public repository via npx during installation and execution phases.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 5, 2026, 05:34 PM
Security Audit — agent-trust-hub — skill-publish