The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest untrusted content from Discord conversations (messages, embeds, and attachments) which may contain malicious instructions.\n
Ingestion points: Data enters the agent's context through the Discord Bot API, user-provided Discord Data Packages, or manually exported local folders as defined in SKILL.md.\n
Boundary markers: The skill includes a 'Trust Boundary' notice and a mandatory staging step in SKILL.md. The instructions explicitly warn the agent not to interpret harvested content as commands. These markers are present to reduce accidental obedience to embedded instructions.\n
Capability inventory: The skill has capabilities to perform network downloads using curl and file system writes (cp, file creation) to save harvested assets, as outlined in references/code-examples.md.\n
Sanitization: Robust sanitization is present; the skill implements sanitize_filename() to prevent path traversal and validate_url() with a strict allowlist and SSRF protections. It also redacts sensitive query parameters from Discord CDN URLs using redact_cdn_url() before logging. Detailed implementations are in references/code-examples.md.

discord-harvest