The Agent Skills Directory

[SAFE]: The skill implements strict credential safety protocols, instructing the agent to never ask users for API keys in chat, print them, or expose them in examples or logs. It correctly identifies that authentication secrets belong solely to the user.
[SAFE]: Software installation is handled securely; the agent is prohibited from automatically installing the CLI and must instead direct the user to official vendor documentation or wait for explicit authorization.
[SAFE]: The skill uses a discovery-based workflow (seli --help) to interact with the system, which ensures the agent operates within the verified capabilities of the locally installed binary rather than guessing API structures.
[INDIRECT_PROMPT_INJECTION]: The skill manages potential injection surfaces from tool output by requiring the agent to parse all CLI responses as structured JSON and verify specific metadata and error fields before acting on data.
Ingestion points: Data enters the agent context via the seli CLI's stdout (found in SKILL.md and references/cli-conventions.md).
Boundary markers: The skill mandates parsing stdout as a single JSON document and checking for an error key before trusting the data payload.
Capability inventory: The agent can execute the seli CLI tool to perform health checks, member lookups, and configuration management.
Sanitization: The instructions require quoting search values, validating JSON shapes before sending via --from-json, and avoiding text-based parsing of command output.

seli-api