The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages workflows that process untrusted data from external sources like GitHub, Slack, and PostHog events.
Ingestion points: Data enters the workflow via internal-event and event triggers defined in SKILL.md.
Boundary markers: The skill instructs the agent to place triggering event data in a separate <triggering_event> block with escaped angle brackets to prevent the data from being interpreted as instructions.
Capability inventory: The created AI tasks have access to repositories, MCP connections, and other skills.
Sanitization: The instructions explicitly warn against unsafe template interpolation of event properties ({event.properties.<name>}) to prevent prompt injection from attacker-controlled event content.
[SAFE]: The skill implements security best practices by requiring actor_access: write filters for GitHub triggers to ensure only authorized users can trigger the automation. It also provides mechanisms to throttle event volume and manage budget through trigger_masking and specific Slack event filters.

building-loops