The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external CMS sources such as WordPress, Webflow, Framer, and Ycode. This introduces a surface for indirect prompt injection where malicious instructions could be embedded in the migrated content to influence subsequent agent actions or the target platform.
Ingestion points: External data is ingested via platform-specific exporters and normalized using utilities in scripts/lib/content-common.mjs and the validation logic in scripts/content-validate.mjs.
Boundary markers: The skill enforces a normalized JSON schema for all collections as defined in references/content-format.md.
Capability inventory: The skill has capabilities to write to the local filesystem (ContentWriter class), perform network downloads (fetch calls in mirror-assets.mjs and seed-content.ts), and interact with databases (Payload CMS SDK in seed-content.ts).
Sanitization: The skill employs a strict allowlist-based HTML sanitizer (sanitizeHtml in content-common.mjs) that strips dangerous tags like <script>, event handlers (on*), and unsafe URL schemes (javascript:, vbscript:) before content is stored or rendered.
[DATA_EXFILTRATION]: The migration process involves downloading media assets from external URLs provided in the source data. To mitigate the risk of secret exposure, the makeClient utility in scripts/lib/content-common.mjs includes a redactUrl function that identifies and redacts credential-like query parameters (e.g., token, key, secret, auth) before URLs are logged or included in export reports.
[COMMAND_EXECUTION]: The skill instructions in SKILL.md guide the user to perform standard development tasks, such as copying scripts and running validation commands (e.g., node tools/migrate/content/content-validate.mjs). These operations are consistent with the tool's primary purpose of content migration and do not represent unauthorized command execution.

migrate-content