kw-stack-audit

Pass

Audited by Gen Agent Trust Hub on Sep 16, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and analyzes various untrusted files from the user's project, which could potentially contain malicious instructions intended to influence the agent's behavior.
  • Ingestion points: Processes package.json, tsconfig.json, CLAUDE.md, and all source files in src/ to count occurrences of as any, @ts-ignore, and console.log.
  • Boundary markers: The instructions do not specify explicit boundary markers or 'ignore' directives for the content of the files being read.
  • Capability inventory: The skill can perform file writes, execute git commands, install packages via npm/pnpm/bun/yarn, and run external CLI tools.
  • Sanitization: No explicit sanitization or validation of the ingested code/config content is mentioned before the agent processes it for metrics.
  • [COMMAND_EXECUTION]: The skill executes several powerful CLI tools and package managers to perform its audit and apply fixes.
  • Evidence: Uses detected package managers (npm, pnpm, bun, yarn) to install dependencies, run build scripts (PM run build), and execute binary tools via npx/pnpm exec/bunx.
  • Evidence: Invokes the Supabase CLI (supabase gen types) and Biome (biome init, biome check) to manage project configuration.
  • [DATA_EXFILTRATION]: The skill identifies and reports sensitive data such as hardcoded secrets and environment variables to the agent's output context.
  • Evidence: Scans for high-entropy strings, private key headers (-----BEGIN), JWTs, and specific secret prefixes (sk_live_, service_role).
  • Evidence: Inspects git history and tracking status for .env files to detect potential leaks.
  • [DYNAMIC_EXECUTION]: The skill executes local utility scripts and determines execution paths at runtime based on the environment.
  • Evidence: Executes a maintenance script located at a computed path in the user's home directory: "$HOME/.claude/gsd-core/bin/gsd-tools.cjs".
  • Evidence: Dynamically determines the appropriate package manager (PM) and execution tool (PMX) by inspecting lockfiles (package-lock.json, pnpm-lock.yaml, etc.).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 16, 2026, 09:14 AM
Security Audit — agent-trust-hub — kw-stack-audit