The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes an automated pipeline that ingests production error logs, which can contain attacker-influenced data, and passes them to an AI system for automated remediation.
Ingestion points: The skill captures server function failures and client-side JavaScript/React crashes which are inherently untrusted sources (SKILL.md).
Boundary markers: The instructions do not define specific delimiters or instructions to help the agent distinguish between logging data and instructions (SKILL.md).
Capability inventory: The workflow includes an 'ai-runner' designed to 'open a fix PR', indicating the system has code-writing capabilities (SKILL.md).
Sanitization: The skill mentions 'redacted at write time' for secrets, but lacks validation or sanitization to prevent adversarial instructions embedded within error messages from influencing the AI runner.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external component from the Convex development ecosystem.
Evidence: The workflow involves installing and configuring the @convex-dev/sentinel package in the user's project configuration (SKILL.md).

convex-sentinel