The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the bash tool to perform file operations (ls, cat, rename) and to execute a specialized utility named gsd-tools. This utility is used for phase management and committing documentation changes within the workspace.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from project files and interpolates it into shell commands, creating a vulnerability surface for indirect injection.
Ingestion points: The skill reads phase descriptions and context from files including .planning/ROADMAP.md, CONTEXT.md, and RESEARCH.md in SKILL.md (Steps 2 and 4).
Boundary markers: There are no explicit delimiters or instructions telling the agent to treat the content of these files as untrusted data rather than instructions.
Capability inventory: The skill possesses the capability to execute arbitrary bash commands and perform file system writes and deletions.
Sanitization: Content from the ${DESCRIPTION} variable, extracted from the ROADMAP file, is passed directly as an argument to gsd-tools query phase.add in Step 4. If a description contains shell metacharacters or escape sequences, it could lead to unintended command execution.

gsd-review-backlog