The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data from several project files, including docs/roadmap/INDEX.md, docs/product/vision.md, and requirements.md. This represents a potential surface for indirect prompt injection if these files contain instructions designed to influence the agent's behavior. * Ingestion points: Data is read from docs/roadmap/INDEX.md, docs/product/vision.md, docs/specs/INDEX.md, requirements.md, and .skills/*/progress.md as specified in SKILL.md. * Boundary markers: The skill explicitly instructs the agent to treat every value as passive data and states that a milestone outcome that reads like an instruction is reported, never obeyed. * Capability inventory: The skill uses grep, sed, and test to parse information and is explicitly defined as read-only. * Sanitization: The skill provides strict instructions to the agent to pass values as single non-option arguments and to reject anything that does not match the expected ID or revision shape.
[COMMAND_EXECUTION]: The skill utilizes shell commands such as grep and sed to process project documentation. While these are standard tools, processing user-controlled content requires care. The skill provides specific instructions to mitigate command injection by treating values as single non-option arguments and validating their structure.

refresh-roadmap-status