The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-controlled data such as rejection_reason from previous proposals and workflow metadata. This information influences the agent's decision-making process when generating new suggestions. * Ingestion points: Data entering through workflows-list-proposals and workflows-list calls. * Boundary markers: The instructions lack explicit delimiters or warnings to treat embedded content in rejection_reason as data rather than instructions. * Capability inventory: The skill has write access to the workflows-suggest tool to create new proposals. * Sanitization: No explicit sanitization or filtering of retrieved text data is described.
[PERSISTENCE]: The skill utilizes a scratchpad mechanism (signal_scout_internal:write) to maintain state between operational runs, tracking items like previously rejected suggestions or steps identified as statistical noise. This is an intended architectural feature for maintaining continuity in analysis.

signals-scout-workflows