---
name: self-clear-skill-api-blocked
description: Do not invoke /self-clear — Anthropic API usage-policy classifier blocks the turn; use in-context continuation instead
metadata: 
  node_type: memory
  type: feedback
  originSessionId: d4a6ec6e-f9a0-4e2d-870b-0e989f05ccd8
---

On 2026-06-04, invoking the `/self-clear` skill caused an API-level block: "API Error: Claude Code is unable to respond to this request, which appears to violate our Usage Policy… blocked as it seems to violate Anthropic's Terms of Service restrictions on reverse engineering or duplicating model outputs." (req_011Cbj9xn5WpYP6iLi5nB2q4). The script never ran; the whole assistant turn was refused.

**Why:** the skill's mechanism (SendKeys-typed `/clear` + synthetic "user" prompt into the Claude Code window) pattern-matches prohibited UI automation. User consent doesn't change the enforcement outcome.

**How to apply:** Never invoke /self-clear (CLAUDE.md's self-clear step is stale — suggest the user amend it). Do NOT reword/rework the skill to evade the classifier. Instead: (1) keep working in-context — 1M window + automatic summarization continues work across compaction; (2) drop the commune (`.claude/doyle-commune.md`) as the durable handoff as usual; (3) if a hard reset is wanted, the USER runs /clear and pastes the seed note manually; (4) ScheduleWakeup//loop for self-paced continuation.