The Agent Skills Directory

[COMMAND_EXECUTION]: Uses the Bash tool to automate planning workflows and validation checks as part of the phase-based execution logic.
[EXTERNAL_DOWNLOADS]: Includes the WebFetch tool to allow the agent to conduct domain research for project phases from external sources.
[PROMPT_INJECTION]: The skill processes external documentation such as PRDs, ADRs, and roadmaps as inputs, which constitutes a surface for indirect prompt injection.
Ingestion points: ROADMAP.md, RESEARCH.md, REVIEWS.md, and any files provided via --prd or --ingest arguments.
Boundary markers: No explicit delimiters or boundary instructions are defined in the skill metadata to isolate ingested data.
Capability inventory: The skill has access to Bash, Write, Agent, and WebFetch tools.
Sanitization: No explicit validation or filtering logic for the ingested file content is specified.
[SAFE]: The skill accesses vendor-specific workflow and reference files located in ~/.claude/gsd-core/. These are established configuration paths for the open-gsd ecosystem and do not indicate unauthorized file access.

gsd-plan-phase