The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to verify code integrity and deployment readiness.
Evidence: npx tsc --noEmit and CONVEX_AGENT_MODE=anonymous npx convex dev --once in SKILL.md.
[DYNAMIC_EXECUTION]: The skill generates application code in the convex/ directory and subsequently uses the Convex CLI to process and deploy that code.
Evidence: The DEGRADATION RULE and SELF-VERIFY RULE describe writing source files and then running the deployment tool.
[EXTERNAL_DOWNLOADS]: The skill references and encourages the use of an external library managed by the Convex organization.
Evidence: Mentions @convex-dev/agent in the rules section of SKILL.md.
[INDIRECT_PROMPT_INJECTION]: The skill translates natural language user requests into executable backend code, which inherently creates an entry point for potential malicious instructions embedded in user data.
Ingestion points: User-provided product ideas and backend requests described in the 'Workflow' section of SKILL.md.
Boundary markers: No explicit prompt delimiters or instruction warnings are specified in the rules.
Capability inventory: The agent can write files to the convex/ directory and execute shell commands (npx).
Sanitization: No explicit input sanitization or filtering logic is provided; the skill relies on Convex platform primitives and standard TypeScript compilation for structural safety.

convex-design