The Agent Skills Directory

[COMMAND_EXECUTION]: The extract.py script utilizes subprocess.run to retrieve metadata and history via git log. These calls are implemented using argument lists without invoking a shell, which effectively prevents shell injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and summarizes untrusted repository content (roadmap, specs, and commit messages).
Ingestion points: Documented in extract.py, which reads from docs/roadmap/INDEX.md, docs/specs/INDEX.md, and Git history output.
Boundary markers: While the source files lack explicit markers, the agent's summarized notes are constrained by instructions to use only names and IDs present in the structured JSON snapshot.
Capability inventory: Limited to local file system reads/writes and Git command execution.
Sanitization: Content is extracted using specific regular expressions, and the render.py script applies strict HTML escaping to all data rendered in the dashboard to prevent cross-site scripting (XSS).
[SAFE]: No network connectivity, obfuscation, or external dependencies were detected. The skill performs deterministic, local operations within the repository's context.

chart-progress