The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources (local logs, deployment subscriptions, production error rows) which could contain adversarial content that might influence agent behavior.
Ingestion points: Data enters the agent context via the wait_for_event tool which returns typed events like convex_error, prod_error, and feature_request (SKILL.md).
Boundary markers: None present; the skill does not instruct the agent to ignore potential instructions embedded within the event payloads.
Capability inventory: The agent is explicitly instructed to "fix" errors and "build" features based on event data, implying capabilities for file system modification and code generation.
Sanitization: No validation or sanitization of the external event data is specified in the skill instructions.

convex-monitor