The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Ycode backup files (.ycode) which contain user-generated CMS content and HTML structure.
Ingestion points: scripts/ycode-export.mjs and scripts/ycode-backup-inventory.mjs read and parse the JSON/Gzip content of backup files.
Boundary markers: The processing is handled by local scripts before being presented to the agent context, though the instructions suggest a workflow where the agent interprets the exported content.
Capability inventory: The provided scripts perform file system operations (read/write) and data normalization. The scripts/lib/content-common.mjs utility also includes a fetch-based HTTP client helper.
Sanitization: The skill implements a robust allowlist-based HTML sanitizer in scripts/lib/content-common.mjs that filters out dangerous tags (e.g., <script>, <style>, <embed>) and potentially malicious event handler attributes.
[DATA_EXFILTRATION]: The skill handles backups containing sensitive credentials (e.g., Anthropic, OpenAI, and Google API keys) and implements specific logic to prevent their accidental exposure.
The SECRET_SETTING_KEY regex in scripts/lib/ycode-backup.mjs identifies sensitive keys like api_key, token, and password to ensure they are not printed to logs or exported to JSON.
The export tool explicitly excludes tables known to contain sensitive integration tokens or personal data, such as app_settings, webhooks, and form_submissions.

migrate-ycode