The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-provided data, including advocate rosters and executive mandates. It implements security best practices by explicitly instructing the agent to treat these inputs as claims rather than facts, requiring 'voluntary-basis evidence' (e.g., an actual opt-in message) before processing, which mitigates the risk of instructions embedded in the input manipulating the agent's behavior.
[DATA_EXPOSURE]: The skill manages advocacy rosters containing social media handles and disclosure records. Data storage is restricted to a local structured format (.ndjson) and is mediated through a controlled proposal process via registry-events.py, preventing unauthorized access or broad exposure of participant data.
[COMMAND_EXECUTION]: All platform-specific interactions and data writes are handled through dedicated internal scripts (registry-events.py, bluesky.py, fediverse.py) rather than arbitrary shell commands. The skill does not perform privileged operations or modify system configurations.
[DYNAMIC_EXECUTION]: While the skill submits data to a registry script at runtime, this follows a predefined protocol (operation: propose) for state management within the platform's 'memory' environment, rather than executing dynamically generated or external code.

advocacy-program-designer