improve-project
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses local development tools like
tsc,eslint, andturboto inspect project configurations, linting rules, and workspace task graphs. These commands are executed to resolve effective settings and ensure project alignment without executing arbitrary or untrusted code. - [EXTERNAL_DOWNLOADS]: The skill performs discovery of package metadata and repository structures by interacting with the npm registry and the GitHub API. It utilizes
npm viewand GitHub's REST API for read-only metadata retrieval (e.g., tags, versions, and peer dependencies). These operations are explicitly scoped to avoid downloading unknown tools or executing lifecycle scripts during the discovery process. - [INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface when processing untrusted repository data such as manifests, lockfiles, and source code.
- Ingestion points: Project configuration and source files are read during the audit and alignment workflows (e.g.,
references/vstack/capabilities/project-alignment/REFERENCE.md). - Boundary markers: The
quality-validation.mdprotocol mandates structured review perspectives and evidence-based self-assessment to prevent the agent from being influenced by instructions embedded within data. - Capability inventory: The agent has capabilities for authorized file modification and executing diagnostic CLI tools.
- Sanitization: All changes are subject to a reduction pass and independent review (critique) to justify every addition and ensure safety and simplicity.
Audit Metadata