The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted external data such as pasted analytics exports, trend reports, and scraped pages, which creates a surface for indirect prompt injection. The instructions include explicit safeguards directing the agent to ignore any embedded instructions that attempt to override system states or commit cadences. \n- Ingestion points: User-pasted analytics exports, trend reports, and scraped web pages. \n- Boundary markers: Natural language instructions in the Instructions section explicitly warning to treat these inputs as untrusted and not allow them to change committed cadences or approve batches. \n- Capability inventory: The skill has the ability to propose changes to the channel registry via the registry-events.py script and write to local memory files. \n- Sanitization: The skill relies on natural language instructions for the agent to filter or ignore instructions embedded within external data. \n- [COMMAND_EXECUTION]: The skill interacts with the local script registry-events.py to propose changes to the channel registry. This is a documented interaction for managing project-specific cadence commitments and is treated as a vendor-managed resource interaction.

social-calendar-builder