The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Shell tool for terminal operations and invokes an external utility named 'gsd-sdk' to query project milestones and analyze roadmaps as described in the objective and context sections.
[EXTERNAL_DOWNLOADS]: The skill incorporates WebFetch and WebSearch tools, enabling it to retrieve content from external network sources during workflow execution.
[DATA_EXFILTRATION]: The skill exposes hardcoded absolute file system paths (e.g., '/Users/patrykattc/...') in the execution_context section, which discloses the user's local directory structure. Furthermore, the combination of file access (reading ROADMAP.md and STATE.md) with network-capable tools (WebFetch) creates a potential exfiltration surface if the agent is compromised.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. Ingestion points: Untrusted data enters the agent context through WebFetch output and user-provided GSD_ARGS. Boundary markers: Absent; there are no instructions provided to the agent to ignore or delimit instructions embedded in external content. Capability inventory: The skill has access to Shell (terminal operations), Write (file modification), and Task (subagent spawning) as defined in the cursor_skill_adapter. Sanitization: Absent; no validation or filtering is performed on ingested data before it influences the autonomous phase execution.

gsd-autonomous