The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/codebase-search.py dynamically downloads and executes the @morphllm/morphmcp package from the npm registry using bunx if a local installation is not found. This dynamic execution of unversioned remote code represents a supply chain risk.
[COMMAND_EXECUTION]: The skill uses subprocess.run to execute mcporter and bunx commands. While it uses a list for arguments to mitigate shell injection, it still triggers external binaries with Natural Language inputs from the query argument.
[DATA_EXFILTRATION]: The skill is designed to send codebase context to the MorphLLM service. While this is the intended functionality, users should be aware that their source code is being transmitted to and processed by an external third-party infrastructure.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted codebase content and processes it via an LLM subagent without explicit boundary markers or sanitization.
Ingestion points: Reads file content from the local repository directory as part of its search process.
Boundary markers: None identified in the script or prompt instructions to isolate codebase content from the agent's core logic.
Capability inventory: The subagent can perform grep and read operations across the entire repository.
Sanitization: No sanitization or filtering of the codebase content is performed before processing by the LLM.

codebase-search