The Agent Skills Directory

[SAFE]: The skill uses a local file-based memory system to store validated content patterns, which is a standard method for maintaining state in agentic workflows.
[SAFE]: All external links point to the author's verified GitHub repository for user support and version history.
[SAFE]: The skill implements a 'Family Pact' (家族公约.md) as a mandatory safety check before processing any data, ensuring that its learning actions remain within predefined ethical and operational boundaries.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted content through user-provided comments and video transcripts. Ingestion points: SKILL.md Phase 1 (Douyin metrics) and Batch Review Step 1 (audience comments and transcripts). Boundary markers: Includes explicit instructions to halt execution if the safety 'Family Pact' (家族公约.md) is unavailable. Capability inventory: Reading and writing local files within the {vault} and aikey/ directories. No shell execution, dynamic compilation, or network operations detected. Sanitization: Employs strict statistical rules and mandatory attribution standards to ensure only verified signals are added to memory.

aikey-retro