The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied input via the $ARGUMENTS variable to determine phase numbers and execution flags. While this creates a potential surface for indirect prompt injection, it is a standard implementation for parameter-driven tasks.
Ingestion points: $ARGUMENTS variable in SKILL.md.
Boundary markers: None explicitly defined for the argument block.
Capability inventory: Access to Bash, Agent, Read, and Write tools.
Sanitization: No specific sanitization logic is visible in the prompt instructions.
[DATA_EXPOSURE]: The skill references specific files in the ~/.claude/gsd-core/ directory as part of its execution context. These paths appear to be internal configuration and reference files for the GSD workflow and are consistent with the skill's stated purpose.

gsd-execute-phase