The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill analyzes local source code and database schema, which serves as an ingestion point for potentially untrusted data during the audit process.
Ingestion points: The agent is instructed to process content from local project files (via grep and git) and database metadata (via SQL queries).
Boundary markers: No specific delimiters or "ignore instructions" warnings are utilized to wrap the ingested content, which could allow instructions hidden within the source code or database to influence the agent.
Capability inventory: The skill utilizes shell commands (grep, git) and database query capabilities to perform its security checks.
Sanitization: There is no explicit sanitization or filtering of the discovered content before it is processed by the agent.
[COMMAND_EXECUTION]: The skill performs automated checks using system commands.
Evidence: Executes git ls-files .env* to verify ignore rules.
Evidence: Executes grep -r "sk_\|key_\|secret" --include="*.ts" --include="*.tsx" to search for hardcoded secrets.
Evidence: Executes SQL queries such as SELECT tablename FROM pg_tables WHERE schemaname='public' AND rowsecurity=false to audit database configuration.

ncode-shipping-checklist