The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes 'brain_frames' checkpointed by headless runs, which include 'retrieved context'. This context is a primary ingestion point for external data that could influence agent behavior.
Ingestion points: Data retrieved from the acc frames command, specifically the 'typed hole' and 'retrieved context' within deliberation frames.
Boundary markers: The instructions lack specific delimiters or boundary markers to isolate potentially adversarial instructions within the retrieved context.
Capability inventory: The skill utilizes the acc_act tool to submit proposals, allowing the processed data to drive actions in the external system.
Sanitization: No sanitization or validation logic is specified for the external context before it is deliberated upon.
[COMMAND_EXECUTION]: The skill requires the use of the acc CLI tool and associated MCP verbs to perform its core functions.
Evidence: Instructions for executing acc frames to list the queue and acc_act to submit calibrated proposals.

accint-frames