The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data from a manifest.json and content.md file, which creates a surface for indirect prompt injection where instructions could be hidden in the inspiration content.
Ingestion points: SKILL.md directs the agent to locate a capture folder and run scripts/validate_capture.mjs on the manifest, and references/thread-brief-contract.md instructs sub-tasks to read the manifest and prompt pack.
Boundary markers: The instructions explicitly define a "Preserve the boundary" section and an "originality boundary" to prevent the agent from copying identity or assets from the source.
Capability inventory: The skill can create new tasks, write code, commit to a repository, and perform private deployments via the Sites tool.
Sanitization: The skill mandates that all identity, copy, metrics, and visuals must be replaced with original content, which acts as a structural mitigation against direct injection from the source material.
[COMMAND_EXECUTION]: The skill executes a local Node.js script to validate the integrity and structure of the inspiration files provided by the user.
Evidence: node <skill-dir>/scripts/validate_capture.mjs <article-dir>/manifest.json is executed during the validation phase.
Impact: The script uses standard Node.js modules (fs, path) to perform filesystem checks (stat and read) to ensure required files like the manifest, content, and images exist and meet specific criteria.

build-daily-inspiration-sites