The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents a surface for indirect prompt injection by fetching content from arbitrary external URLs (changelogs, feeds, YouTube descriptions, App Store listings) for AI-driven processing and summarization.
Ingestion points: The skill uses manage_source (action: "fetch") and the releases admin source fetch command to ingest data from external URLs.
Boundary markers: The instructions do not define specific delimiters or "ignore" guards to prevent instructions embedded in fetched changelogs from influencing the agent's behavior, although it mentions "summarizer-cleaned" data.
Capability inventory: The agent has extensive administrative access via the releases CLI to modify organization data, update playbooks (which contain persistent instructions), and manage sources.
Sanitization: The skill mentions filtering and cleaning but does not detail prompt-level sanitization for preventing injection in the ingested text.
[COMMAND_EXECUTION]: The skill relies on a local command-line interface tool named releases to perform administrative tasks, including organization management and source deletion.
[DATA_EXFILTRATION]: The skill performs network requests to non-whitelisted external domains as part of its core functionality for fetching changelog data and evaluating discovery URLs.

managing-sources