The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and iterative processing of external, potentially untrusted data through a multi-step deliberation cycle.
Ingestion points: External content enters the agent's context through the input parameter in the acc_act(runtime="solve") call and through the intermediate brain_frame data returned by the tool.
Boundary markers: The instructions do not specify any boundary markers, delimiters, or explicit warnings to ignore embedded instructions within the processed data.
Capability inventory: The skill leverages the acc_act tool to execute actions, submit deliberated proposals, and report outcomes based on external inputs.
Sanitization: There is no documentation of sanitization, escaping, or validation of the input goals or the returned frame content before they are used to influence subsequent reasoning steps or tool invocations.

accint-solve