The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data (such as deployment status, user signups, or CI logs) that is subsequently processed by an AI-driven service ("Poke"). This creates a potential surface for indirect prompt injection where untrusted external data could attempt to influence the agent's behavior.
Ingestion points: External data is ingested from HTTP request bodies in references/vercel-pattern.md and from GitHub environment context in references/github-actions-pattern.md.
Boundary markers: The skill instructs users in references/condition-action-model.md to define explicit condition and action strings, which serve as an instructional lens to scope how the agent interprets incoming data.
Capability inventory: The skill utilizes the poke Node.js SDK (referenced in references/vercel-pattern.md) and curl commands (referenced in references/github-actions-pattern.md) to perform outbound network operations to the Poke service.
Sanitization: The skill provides a dedicated guide in references/payload-shape.md that mandates sanitization practices, including the truncation of logs to 500 characters, the use of stable identifiers for deduplication, and a strict prohibition against including secrets or PII in payloads.

poke-webhooks