The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes user-supplied influencer data and campaign context, creating a potential surface for indirect prompt injection.
Ingestion points: Influencer handles, niche descriptions, and deliverables provided by the user, as well as roster records retrieved from memory/creators/<handle-slug>.md (SKILL.md).
Boundary markers: The skill utilizes structured templates in references/templates.md and mandatory 'cold-copy rules' in references/cold-copy-rules.md to constrain agent output, though it lacks explicit instructions to ignore embedded commands within the input data.
Capability inventory: Capability to write to the local memory/ directory and execute the registry-events.py script for registry updates (SKILL.md).
Sanitization: Content is reviewed through an 'expert-panel' persona method, but no formal programmatic sanitization or escaping of ingested data is described.
[COMMAND_EXECUTION]: The skill instructions direct the agent to invoke a local script named registry-events.py using an operation: propose request to update registry events. This interaction is restricted to internal state management and does not facilitate the execution of arbitrary shell commands or untrusted code.
[DATA_EXFILTRATION]: The skill manages potentially sensitive creator data, including historical rates and contact paths, strictly within the local file system (memory/ directory). All network-related data sources are represented by ~~ placeholders, ensuring that no data is transmitted to external servers without explicit user configuration of connectors.

outreach-manager