meta-config-stack-detect
Pass
Audited by Gen Agent Trust Hub on Sep 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill analyzes a wide range of project-specific files (e.g.,
package.json,pnpm-lock.yaml,turbo.json,wrangler.toml, and framework-specific configs) to detect a technology stack. This ingestion of untrusted local data creates a surface where malicious instructions or deceptive metadata embedded in these files could potentially influence the agent's analysis or the content of the generated human-readable report. - Ingestion points: Multiple manifest, lock, and configuration files across workspaces in a user repository (e.g.,
package.json,vite.config.ts,turbo.json). - Boundary markers: The skill does not explicitly use delimiters or specialized instructions to ignore potential injections within the processed files, relying instead on its internal mapping logic and judgment patterns.
- Capability inventory: The skill can read local files, execute vendor CLI tools, and generate shell commands for the user to execute.
- Sanitization: The agent is instructed to ensure the
SeedPayloadremains schema-pure, providing a structural defense against arbitrary data inclusion in the machine-readable output, though the human-readable report remains a surface for potential injection influence. - [COMMAND_EXECUTION]: The skill is instructed to execute shell commands to interact with the vendor's CLI tool (
agents-inc search --help) to verify support for specific output formats and to acquire the necessary data catalogs for mapping discovered libraries to skill IDs. - [DATA_EXFILTRATION]: The skill facilitates the sharing of project metadata (the
SeedPayload) by generating a command for the user to run (agents-inc share --stdin). This process transmits information about the project's structure and dependencies to the vendor's infrastructure for the purpose of generating a shareable configuration ID.
Audit Metadata