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...
- **Before large tasks**: check before any operation that will generate many tool calls or long output
- **Proactively**: if responses feel slower, less coherent, or you notice unusual hedging, run this check immediately
- **After a branch merge or plan execution**: context spikes are common at transition points
π Tips & Best Practices
prompt_count is a proxy, not a direct token count. Actual context consumption varies by response length. Treat thresholds as conservative triggers, not hard limits.
Both state files (anchor_state.json, session_anchor.json) are written by the Willow server. If either file is missing, the script outputs STATUS_OK and logs a warning to stderr β it fails open, not closed.
This skill does not modify any state files. It is read-only and safe to run at any time.
