The Agent Skills Directory

[SAFE]: The skill provides a structured way for agents to track project entities, dependencies, and their purposes. All operations are local to the repository.
[COMMAND_EXECUTION]: The provided script scripts/dsp-cli.py manages the graph metadata. It uses standard file operations (pathlib, shutil) to maintain the .dsp/ directory. No arbitrary command execution or shell injection vectors were found.
[DATA_EXFILTRATION]: No network access or data exfiltration patterns are present. The skill does not communicate with external endpoints and does not harvest credentials.
[EXTERNAL_DOWNLOADS]: The skill is self-contained with no external package requirements or remote script downloads.
[PROMPT_INJECTION]: The agent instructions in SKILL.md are focused on the technical requirements of maintaining the DSP graph and do not attempt to override system safety protocols.
[SAFE]: The skill involves processing potentially untrusted project data (Category 8 surface). Ingestion points: User-provided CLI arguments for entity descriptions (source, purpose). Boundary markers: None. Capability inventory: Local file-system operations (read, write, delete) within the .dsp/ directory. Sanitization: No explicit escaping or validation. The risk is assessed as SAFE given the restricted local scope and documented use case.

data-structure-protocol