The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to invoke multiple external AI command-line interfaces, including claude, codex, opencode, qwen, cursor, and antigravity (or agy). This involves executing third-party binaries that must be present on the host system.
[INDIRECT_PROMPT_INJECTION]: The instructions specify that the output from these external AI CLIs (collected in REVIEWS.md) is designed to be fed back into the agent's planning process via the /gsd-plan-phase --reviews command. This creates a vector where instructions or adversarial content returned by an external service could influence the agent's subsequent planning and execution.
Ingestion points: External CLI responses collected from standard output and written to REVIEWS.md (SKILL.md).
Boundary markers: None specified to distinguish external AI feedback from trusted internal instructions.
Capability inventory: The skill utilizes Read, Write, Bash, Glob, and Grep tools, providing a broad set of capabilities for the agent to act upon potentially untrusted feedback.
Sanitization: There is no evidence of sanitization, validation, or filtering of the content received from the external AI services before it is used in the planning workflow.

gsd-review