The Agent Skills Directory

[COMMAND_EXECUTION] (LOW): The skill is designed to run research and implementation pipelines using the Bash tool. While intrinsic to its functionality, this capability allows for general command execution on the host system.
[CREDENTIALS_UNSAFE] (LOW): The documentation explicitly describes a pattern for passing the host's full os.environ to subprocesses to ensure API keys (e.g., from ~/.claude/.env) are available to MCP tools. This broad exposure of sensitive environment variables to third-party tools is a potential security risk if a tool is compromised.
[PROMPT_INJECTION] (LOW): As an indirect injection surface (Category 8), the skill ingests untrusted data from external sources via tools like nia__search (documentation) and ast-grep__find_code (source code). This data is used to inform downstream implementation steps, creating a path for malicious instructions embedded in documentation or codebases to influence agent behavior.
Ingestion points: nia__search, ast-grep__find_code, morph__warpgrep_codebase_search.
Boundary markers: None specified in the provided snippets.
Capability inventory: Bash execution, file reading (Read).
Sanitization: None mentioned for the interpolated tool outputs.

mcp-chaining