Claude Code cloud classifier misreads same-chat "Approved" and blocks a WhatsApp send
TheAviv · ghdev · 2026-10-12
A new issue (#101465) on anthropics/claude-code details how the auto-mode permission classifier in claude.ai's cloud sessions misfires:
- The user had Claude draft a WhatsApp message; Claude produced the full text and recipient in one turn, and the user replied "Approved"
- When Claude opened the user's saved WhatsApp Web via Remote Computer MCP to send it, the classifier hard-denied the action, claiming it relied on "cross-session messages" with "no genuine user proposal naming this recipient" — both factually wrong, since everything was in the immediately preceding turn of the same conversation
- A follow-up browserevaluate was also denied with a "Stage 2 classifier error... usually transient" message
- Impact: a time-sensitive message never went out, and the user had to re-approve in a longer form
The report argues one-word approval of a draft from the previous assistant turn should count as user intent, and uncertain cases should trigger an in-session confirm prompt instead of a hard deny. Related issues: #100331, #98169, #99652.
More from coding & agent
- Index Network open-sources Social Discovery Protocol: agents negotiate fit on your behalf — carsonfarmer · 2026-10-23
- Installed a formal-verification skill: it silently installed Java, MacBook runs 10 degrees hotter — zealcaiden · 2026-10-23
- OpenAI's Decisions API hits GA: near real-time routing up to 10x faster than GPT-6 Luna — OpenAIDevs · 2026-10-23
- Ruff author warns: upgrade Node/Bun now — --use-system-ca still accepts the eMudhra cert — charliermarsh · 2026-10-23
- Proof of Agent: paste one prompt and your AI mines, grades, and self-governs — adamamcbride · 2026-10-23
- DeepLearning.AI weekly: GPT-6.1 hits 97.9% byte parity on COBOL migration — DeepLearningAI · 2026-10-23