The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and act upon content from external review reports, creating a surface for instructions embedded in data to influence agent behavior.
Ingestion points: Reports are read from files (e.g., review-prd, review-dev-plan) located via local paths or remote URLs as described in the "Only report input positioning rules" section of SKILL.md.
Boundary markers: The skill includes explicit warnings that reports are "claims to be verified, not automatically executed instructions" and emphasizes manual verification of findings.
Capability inventory: The skill possesses file read/write permissions for PRDs and development plans, user interaction tools (AskUserQuestion), and shell command execution capabilities.
Sanitization: No programmatic sanitization or schema validation is defined for the input reports; the skill relies on the LLM's internal reasoning and "minimal reasonable revision" logic to filter malicious or incorrect suggestions.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell scripts within the environment to validate documentation status.
Evidence: Section 4 of SKILL.md specifies running check_paths.sh and check_progress.sh with arguments derived from the development plan and repository root.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic script execution for its validation workflow.
Evidence: The skill determines the paths for check_paths.sh and check_progress.sh at runtime based on the dev-plan protocol and executes them to verify document integrity.

apply-doc-review