The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted documentation and task files from the project repository. \n- Ingestion points: The agent reads various project files (PRDs, task lists, READMEs) to perform audits and migrations (defined in SKILL.md and governance-migration.md). \n- Boundary markers: The skill instructs the agent to 'Read only the references you need' and 'Determine the action and request boundaries' before acting. \n- Capability inventory: The agent has permissions to read/write Markdown files and propose decisions. It can also execute optional project-local scripts if present. \n- Sanitization: The skill relies on 'Decision approval' rules, requiring explicit user acceptance for any material change or choice, which serves as a manual check against unintended obedience to embedded instructions. \n- [DYNAMIC_EXECUTION]: The validator-compatibility.md reference describes the use of optional Node.js validation scripts (e.g., node scripts/check-docs.mjs). The documentation explicitly states that these scripts are not shipped with the skill and must be provided by the project environment. This pattern is for compatibility with existing user-controlled CI/CD workflows rather than arbitrary code execution by the skill itself. \n- [SAFE]: The skill consists entirely of human-readable Markdown documentation and instructional templates. Analysis of all files confirmed the absence of obfuscation (Category 3), hardcoded credentials (Category 2), or unauthorized privilege escalation (Category 5).

dev-docs