The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes a shell-based validation section for verifying character counts and git status.
Evidence: SKILL.md defines a 'Validation steps' section utilizing wc -c, grep, and git add to ensure output files meet specific constraints and are staged locally. These operations are restricted to the local project environment.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection as it ingests untrusted project files to generate new instructions.
Ingestion points: Reads architecture documents, previous goals, git logs, and source code (SKILL.md).
Boundary markers: Not present.
Capability inventory: Read, Write, and Bash execution capabilities are requested by the skill.
Sanitization: The skill does not explicitly sanitize the content of the read files before using them in the generation process, though it instructs the agent to 'verify they match HEAD before citing'.
[SAFE]: The skill explicitly forbids dangerous behaviors such as git push or arbitrary architectural changes without logging them in a candidate file. It prioritizes local, phased commits and adherence to project-agnostic documentation standards. All external links target legitimate author documentation, license text, or well-known services.

goal-rider-author