The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input from a public web form (e.g., first name, last name, and message fields) which is later ingested by an AI lead researcher agent for analysis.
Ingestion points: Public form fields defined in infra/tools/inbound-form.ts and site/components/inbound-form.tsx.
Boundary markers: The agent in infra/agents/lead-researcher.ts uses a structured system prompt and a specific JSON output schema to constrain model behavior and prevent hallucination or instruction override.
Capability inventory: The researcher agent utilizes webSearch and context (read-only). The associated workflows perform postMessage (Slack) and modelCustomColumn (writes to GTM models).
Sanitization: Input fields are validated using a Zod schema in the inbound_form workflow definition to ensure data integrity before processing.
[EXTERNAL_DOWNLOADS]: The skill utilizes vendor-specific libraries for deployment, infrastructure definition, and frontend form management.
Evidence: References to @cargo-ai/cli, @cargo-ai/cdk, and @cargo-ai/form-sdk in SKILL.md, evals/contract.mjs, and references/form.md.
[COMMAND_EXECUTION]: The skill instructions and evaluation scripts utilize the vendor's CLI tool for administrative and deployment tasks.
Evidence: Usage of cargo-ai cdk plan, cargo-ai cdk deploy, and cargo-ai storage column list as documented in the skill's operational procedures.

inbound-qualification