clear-contextlisted

## What This Does Implements the decision tree from rule 014 (`cf-tool-result-clearing`) as an executable workflow. Measures accumulated tool result and thinking block size, picks the cheapest documented primitive, and emits the required `context_management` params. **Decision tree:** ``` Tool results > 30K AND results are re-fetchable? YES → use clear_tool_uses_20250919 NO → thinking blocks stale or too large? YES → use clear_thinking_20251015 NO → consider documented compaction guidance outside this skill ``` ## When to Use **Trigger:** Tool results accumulating past 30K tokens; stale or excessive thinking blocks; conversation approaching 50% context window; before spawning expensive Tier 3 subagent; proactively every 20+ tool calls. **Never needed:** When context is fresh (<10K tool result tokens). ## How to Use ### Step 1 — Measure tool result accumulation Estimate by counting recent tool result blocks. Rough formula: - Each file read ≈ (file lines × 5) tokens - Each rg result ≈ (match count × 20) tokens - Each bash output ≈ (output lines × 8) tokens If estimate > 30K: proceed to clearing. ### Step 2 — Determine if results are re-fetchable Re-fetchable (safe to clear): file reads, search results, directory listings, git status. NOT re-fetchable (do NOT clear): agent responses containing decisions, architectural choices, error root causes, any content not reproducible by re-running the tool. ### Step 3 — Emit context_management params **F