The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes local project manifest files (such as package.json, pyproject.toml, go.mod, and Cargo.toml) to detect existing configurations and dependency states. This ingestion of external data constitutes a theoretical attack surface for indirect prompt injection if the project being audited contains malicious instructions hidden within its metadata or dependency fields.
Ingestion points: The skill reads manifest and lockfiles across JS/TS, Python, Go, and Rust projects.
Boundary markers: The instructions do not define explicit delimiters to isolate the content of ingested manifests from the agent's logic, though the skill's operations are strictly focused on specific configuration transformations.
Capability inventory: The skill possesses the capability to modify local project files (Write, Edit) and execute ecosystem-specific commands (Bash via npm, pnpm, pip, go, cargo) to apply hardening measures.
Sanitization: The risk is mitigated by the skill's use of standardized configuration templates and the highly structured nature of the data it processes (versions and dependency names). As the ingestion of these files is essential to the skill's primary defensive function, the risk is categorized as safe.

deps