The Agent Skills Directory

[SAFE]: The skill follows a structured problem-solving methodology without executing dangerous system commands, performing unauthorized network operations, or hardcoding sensitive credentials. The instructions focus on logical classification and routing.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it processes external data to determine execution paths.
Ingestion points: Untrusted data enters the agent context through the $ARGUMENTS variable in SKILL.md and through user responses collected via the AskUserQuestion tool.
Boundary markers: The skill implements procedural boundary markers, including specific instructions to check for blank responses from the AskUserQuestion tool and a requirement to challenge the user's initial framing (Adjacent Domain Challenge) before confirming a classification.
Capability inventory: The skill's capabilities are limited to calling the AskUserQuestion tool and routing to other predefined skills; it does not contain direct file-write, network exfiltration, or arbitrary code execution logic.
Sanitization: External content is sanitized through the Cynefin triangulation logic, which requires mapping inputs to specific technical domains and verbs, acting as a structural filter for the agent's behavior.

frame-problem