@rudi193-cmd

willow-external-guard wraps untrusted external content (web, jeles, corpus, sub-agent) in sandwich defense markers and scans for prompt injection, role hijack, and leak attacks before kb writes or llm passes. response levels (warn, confirm, block) are context-dependent on source type and threat confidence.

willow-memory-health audits openclaw agent memory files for staleness, redundancy, dark records, and contradictions. runs a diagnostic script against the memory directory, interprets buckets (hot/warm/stale/dead), and offers guided cleanup actions with confirmation checkpoints.

Audit the Willow local AI stack for subsystem failures, drift, and resource bloat. Use when a user asks to check Willow health, diagnose a slow or broken Wil...

Use when you need to check whether the current session is approaching context limits and decide whether to compact, hand off, or continue. Implements a casca...