The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the system using subprocess.run and os.execve to execute commands for Git repository discovery, lsof process auditing, and interaction with the codebase-memory-mcp CLI. These operations are performed with list-based arguments to prevent shell injection and include validation of executable paths.
[DYNAMIC_EXECUTION]: The shell script scripts/codebase-memory-graph.sh utilizes Python heredocs (`python3
<<'PY'`) to perform complex JSON processing and path resolution. This logic is statically defined within the script and used for local data transformation.
[INDIRECT_PROMPT_INJECTION]: The skill processes repository metadata and project graph structures which could be influenced by untrusted data in a target repository. It mitigates this risk through the use of robust parsing techniques, such as Git's porcelain output formats with NUL-separators and comprehensive validation of all ingested project paths.
Ingestion points: Repository structures analyzed via git in scripts/codebase_memory_cache.py and scripts/codebase-memory-graph.sh, as well as JSON manifest files.
Boundary markers: The skill enforces strict JSON schemas and utilizes NUL-separated command output (e.g., git worktree list -z) to ensure data integrity during parsing.
Capability inventory: File system operations (reading/writing manifests and database files), local network status checks (via curl to localhost), and management of the MCP service CLI.
Sanitization: Implements extensive checks for NUL characters in paths, verifies that paths are absolute and within expected boundaries, validates manifest SHA-256 digests, and uses restricted file permissions (chmod 600) for sensitive temporary data.

codebase-memory-mcp