The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes an indirect prompt injection surface by reading content from the repository (README, manifest files, and source code comments) to infer project tasks. While this allows malicious instructions in the scanned files to potentially influence the agent's output, the impact is low as the functionality is limited to generating documentation.
Ingestion points: Scans README.md, package.json, pyproject.toml, Cargo.toml, and various source code files.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are used for the ingested content.
Capability inventory: The skill is capable of writing to the local filesystem using a 'Write tool'.
Sanitization: The skill does not perform explicit validation or sanitization of the scanned file content before processing it.
[COMMAND_EXECUTION]: The skill performs repository scans, greps for specific developer markers (TODO, FIXME, HACK), and writes a ROADMAP.md file. These file-system operations are restricted to project documentation and metadata, which is consistent with the skill's primary purpose.

roadmap-init