The Agent Skills Directory

[PROMPT_INJECTION]: The skill utilizes untrusted data from the local repository to ground its actions.
Ingestion points: Reads files such as package.json, AGENTS.md, README.md, and git log to establish project context.
Boundary markers: The skill advocates for validation at system boundaries (Rule R8) but does not define specific separators for interpreting repository content.
Capability inventory: Possesses capabilities to modify the file system, execute shell commands (linting, formatting), and perform database migrations.
Sanitization: Emphasizes internal assertions and boundary validation as architectural principles.
[EXTERNAL_DOWNLOADS]: The skill references reputable external libraries and educational resources.
Dependencies: Recommends standard, widely-used packages like zod, stripe, and ioredis.
Educational Resources: References established GitHub repositories for system design and algorithms (e.g., donnemartin/system-design-primer, trekhleb/javascript-algorithms).
Trusted Vendors: Leverages official documentation and tools from trusted entities including Anthropic, Microsoft, and Vercel.
[COMMAND_EXECUTION]: Utilizes common development tools to maintain code quality.
Quality Gates: Employs standard commands for eslint, prettier, and tsc to enforce style and type safety.
Safety Measures: Explicitly instructs the agent to seek operator approval before executing destructive operations like rm -rf or database deletions.
[DATA_EXFILTRATION]: The skill mandates safe secret management.
Best Practices: Forbids committing secrets, recommends .env.example templates, and ensures sensitive configuration is ignored by version control.
PII Protection: Prohibits logging personally identifiable information or internal database details.

production-grade