The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes a system where user-provided inputs from the POST /kickoff endpoint are interpolated into LLM prompts using {placeholder} tokens.
Ingestion points: Data enters via the inputs dictionary in the POST /kickoff request body (SKILL.md, references/api-contract.md).
Boundary markers: The skill documentation notes that CrewAI uses curly braces for interpolation and warns that literal braces in prompts (e.g., JSON samples) can be misidentified as input placeholders (SKILL.md, Section 2).
Capability inventory: The skill facilitates the execution of AI agents and tasks which may involve LLM calls and tool execution (SKILL.md, Section 8).
Sanitization: The skill recommends validating input keys against the GET /inputs endpoint before kickoff and notes that input values are stringified using str() before interpolation (SKILL.md, Section 2).
[EXTERNAL_DOWNLOADS]: The skill reference implementation depends on standard, well-known libraries and tools.
Evidence: The Python client requires httpx, and the documentation mentions crewai and jq for the bash implementation (references/python-client.md).

call-deployed-crew