directions-lab

Warn

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill uses dynamic module imports (import()) on paths constructed at runtime. This behavior is found in scripts/lib/capabilities.mjs, which loads a feature index script from the project root, and in scripts/lib/render.mjs, which dynamically loads various file templates during the scaffolding process.
  • [COMMAND_EXECUTION]: The scaffolding logic in scripts/scaffold.mjs and scripts/lib/verify.mjs executes system commands via child_process.spawnSync and execFileSync. These are used to verify the generated lab by running vitest, vite build, and pnpm typecheck. The skill also generates shell command configurations for user launch profiles in .claude/launch.json.
  • [INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests user input via CLI flags and interpolates it into generated source code.
  • Ingestion points: Flag inputs in scripts/scaffold.mjs such as --slug, --title, and --surfaces.
  • Boundary markers: None identified in the templates where flags are interpolated.
  • Capability inventory: Execution of shell commands via spawnSync in scripts/lib/verify.mjs and file system modifications/deletions in scripts/scaffold.mjs.
  • Sanitization: Slugs and IDs are validated with regular expressions, but descriptive labels and names used in templates receive less stringent filtering.
  • [EXTERNAL_DOWNLOADS]: The skill instructions in SKILL.md advise running pnpm install to manage dependencies. Additionally, the capture utility assets/templates/shot.cjs.mjs relies on the external playwright package and its browser binaries to generate screenshots.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 16, 2026, 12:30 AM