lisa-update-projects

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the latest version of the vendor package @codyswann/lisa from the npm or Bun registry during the project update process.
  • [COMMAND_EXECUTION]: Executes shell commands including git, npm, bun, node, and jq to modify project files, manage dependencies, and handle version control operations.
  • [REMOTE_CODE_EXECUTION]: Runs vendor-provided JavaScript files, specifically node_modules/@codyswann/lisa/dist/index.js and scripts/detect-stale-workflow-inputs.mjs, to apply templates and validate workflow configurations.
  • [PROMPT_INJECTION]: The skill processes untrusted local data which creates a surface for indirect prompt injection.
  • Ingestion points: Reads project configurations from .lisa.config.local.json, package.json, and .Codex/settings.json, as well as file content from .github/workflows/.
  • Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized when interpolating data from these files into agent logic.
  • Capability inventory: The agent has extensive shell execution capabilities (git, bun, npm, node) and file-system write access across the project directory.
  • Sanitization: Input data from local configuration files is not explicitly sanitized before being used to determine branch names, package manager commands, or workflow mappings.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 06:27 AM
Security Audit — agent-trust-hub — lisa-update-projects