The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill follows security best practices for secret management. It instructs the AI agent to read the vendor API key from an environment variable and strictly forbids the agent from saving, logging, or echoing the key in any command output or generated file.
[COMMAND_EXECUTION]: All API interactions are gated by a specialized Python script (scripts/ingfah_api.py). This script enforces a rigid allowlist of permitted API routes and requires an explicit confirmation flag (--confirm) before any state-changing operations (POST, PUT, DELETE) are executed, ensuring user oversight for real-world effects like placing outbound calls.
[EXTERNAL_DOWNLOADS]: The skill interacts exclusively with trusted vendor domains (api.ingfah.ai and docs.ingfah.ai) for its core functionality. These domains are associated with the official Ingfah platform and are used to fetch documentation and manage agent resources.
[INDIRECT_PROMPT_INJECTION]: While the skill processes potentially untrusted data like call transcripts and knowledge base files, it provides the agent with explicit defensive instructions. These rules direct the agent to prioritize its primary mission and safety guidelines over any instructions found within the ingested data.
[PROMPT_INJECTION]: The skill includes dedicated sections on 'Prompt security' and 'AI disclosure,' instructing the agent never to reveal its underlying instructions or comply with override attempts like 'ignore previous instructions.'

ingfah