adapt

Pass

Audited by Gen Agent Trust Hub on Sep 2, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it ingests and processes untrusted data from various project files to generate instructions for the user and updates to configuration files.
  • Ingestion points: Reads CLAUDE.md, package.json, Package.swift, Cargo.toml, pyproject.toml, go.mod, Gemfile, plugin.json, jest.config.*, vitest.config.*, pytest.ini, and CI/CD workflow files.
  • Boundary markers: No explicit delimitation or instructions to ignore embedded commands are present when processing these files.
  • Capability inventory: The agent is capable of writing to the local file system (CLAUDE.md) and creating directories (docs/superpowers/specs/, docs/superpowers/plans/).
  • Sanitization: No sanitization or validation of the content read from the project files is described before it is used to summarize project state or update the routing logic.
  • [EXTERNAL_DOWNLOADS]: The skill provides instructions for the user to manually clone and execute a setup script from an external repository (github.com/garrytan/gstack.git) if the dependency is missing. This reference is part of the skill's primary setup function.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 2, 2026, 02:34 AM
Security Audit — agent-trust-hub — adapt