gardener
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources such as package manifests, lockfiles, and git logs from peer repositories to inform its maintenance decisions.
- Ingestion points: Data is pulled into the agent context via npm registry queries (
pnpm view), peer repository mirrors in.agent_sources/, and generated metadata in.brain/data/unlinked-mentions.generated.json. - Boundary markers: There are no explicit delimiters or instructions provided to ignore potentially malicious embedded text within the external manifests or logs.
- Capability inventory: The skill possesses significant capabilities, including shell command execution (
pnpm,git,sh), filesystem write access to configuration files (package.json), and the ability to install new software packages. - Sanitization: No specific sanitization or filtering logic is mentioned for the content retrieved from external peer repositories before it is processed by the agent.
- [EXTERNAL_DOWNLOADS]: The skill fetches package metadata and installs software from the npm registry.
- Evidence: Use of
pnpm viewto query registry tags andpnpm installto download dependencies. These operations target established, well-known package registries. - [COMMAND_EXECUTION]: The skill relies on several shell-based tools to perform repository audits and updates.
- Evidence: Extensive use of
pnpmfor management,gitfor history inspection, andrg(ripgrep) for searching codebase patterns. - [DYNAMIC_EXECUTION]: The skill automatically triggers the project's build and test suites to verify updates.
- Evidence: Execution of
pnpm turbo run check test build --force, which runs project-defined scripts and tests as part of the validation process.
Audit Metadata