dereks-agentic-skills

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSOBFUSCATION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The PR quiz system in philipp-android-skills/pr-quiz/SKILL.md reads GitHub Pull Request diffs using gh pr diff. This creates an ingestion point for untrusted data that could potentially contain malicious instructions intended to influence the agent's behavior or the generated quiz content.
  • Ingestion points: philipp-android-skills/pr-quiz/SKILL.md (via gh pr diff).
  • Boundary markers: Employs <!-- pr-quiz:start --> and <!-- pr-quiz:end --> delimiters for quiz blocks.
  • Capability inventory: The publish_quiz.py script uses subprocess.run to call gh and edit pull request descriptions.
  • Sanitization: The one_line function in publish_quiz.py normalizes input strings to prevent multi-line injection attacks.
  • [COMMAND_EXECUTION]: The skill contains several scripts that execute shell commands for project management and environment setup.
  • philipp-android-skills/pr-quiz/scripts/publish_quiz.py uses subprocess.run to execute GitHub CLI (gh) commands to view and edit pull requests.
  • android-design-gallery-and-signing/scripts/check-signing-safety.sh executes keytool for keystore generation and git commands for security auditing.
  • multi-agent-orchestration/scripts/detect-env.sh probes for installed tools like adb, scrcpy, and orca using command -v.
  • [EXTERNAL_DOWNLOADS]: The documentation references installation scripts from a well-known service for setup.
  • philipp-android-skills/README.md instructs users to install Claude Code using curl or irm commands targeting claude.ai (official Anthropic service).
  • [OBFUSCATION]: The PR quiz grading system uses Base64 encoding to store metadata within pull requests.
  • philipp-android-skills/pr-quiz/scripts/publish_quiz.py encodes a JSON object containing the quiz answer key and explanations into a Base64 string, which is then placed in a hidden HTML comment in the PR description. This is used for verification by the corresponding GitHub Action.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 10:12 PM