The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads source code from well-known package registries including npm, PyPI, and crates.io, as well as GitHub repositories. These downloads are directed to a global cache at ~/.opensrc/.\n- [COMMAND_EXECUTION]: The skill uses the opensrc CLI tool along with standard shell utilities like rg, cat, and find via the Bash tool. This allows the agent to execute commands to search for and read implementation details within the fetched source code.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it encourages the agent to read and analyze source code from external repositories. Maliciously crafted instructions hidden in code comments or documentation in those repositories could potentially manipulate the agent's actions.\n
Ingestion points: External code and documentation are ingested into the agent's context when it reads the files fetched to the ~/.opensrc/ directory.\n
Boundary markers: There are no explicit delimiters or system instructions provided to help the agent distinguish between its own guidelines and instructions that might be embedded in the external source code.\n
Capability inventory: The agent has access to the Bash tool, which provides a broad set of capabilities that could be exploited if an indirect injection attack is successful.\n
Sanitization: The skill does not include any mechanism to sanitize or validate the content of the external repositories before it is processed by the agent.

opensrc