The Agent Skills Directory

[SAFE]: The skill's logic and instructions do not contain malicious code, obfuscation, or unauthorized external interactions. It functions as a coordinator for project workflows.
[COMMAND_EXECUTION]: The skill executes local scripts such as gsd-tools.cjs and uses agent-spawning tools to manage tasks. These operations are performed within the local workspace and are necessary for the skill's stated purpose of phase management.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface through its ingestion of local planning files.
Ingestion points: Reads contents of .planning/STATE.md and .planning/ROADMAP.md during the workflow refresh loop.
Boundary markers: No explicit delimiters are specified in the wrapper instructions for the content of these files.
Capability inventory: Features the ability to spawn sub-agents (spawn_agent), execute skill commands, and write to user terminals.
Sanitization: Content is processed by a workflow engine that utilizes structured mapping for agent collaboration.
Evaluation: The risk is minimal as the ingested files are local project metadata, and the execution relies on structured agent interfaces rather than direct shell interpolation.

gsd-manager