The Agent Skills Directory

[SAFE]: The skill operates as a read-only analysis tool, accessing only local documentation files (roadmap.md and decisions.md) within specific project subdirectories. It does not perform any network operations or access sensitive system configuration files.
[COMMAND_EXECUTION]: The skill utilizes a built-in platform command /emit-pending-review to record analysis findings. This command is restricted to creating structured review files in a designated .pending-reviews/ folder and does not allow for arbitrary code or shell command execution.
[PROMPT_INJECTION]: The skill processes untrusted input from local roadmap files, creating a potential surface for indirect prompt injection.
Ingestion points: Data is read from roadmap.md and decisions.md as specified in SKILL.md.
Boundary markers: The instructions define clear behavioral boundaries (read-only, no editing), although no specific syntactic delimiters are applied to the ingested content.
Capability inventory: The skill's only write capability is via the structured /emit-pending-review tool; it has no access to network or subprocess execution.
Sanitization: The skill does not sanitize or escape the content of the analyzed files before processing.

review-roadmap