Summary

skill_workshop action=apply consistently returns {"status":"error","error":"Approval timed out"} for a large proposal. It does not partially apply (proposal stays pending, live SKILL.md unchanged), so retrying is safe — but the apply never completes through the tool.

Environment

• OpenClaw 2026.6.8 (844f405)
• node v22.22.0, macOS (arm64)

Repro

1. Create/update a skill via skill_workshop so the proposal's PROPOSAL.md is large — in our case 27,986 bytes (a workspace skill, client-monitoring).
2. Call skill_workshop action=apply with the proposal id.
3. Returns error: "Approval timed out". Repeated 3× consecutively — consistent, not transient.

Expected

Apply writes PROPOSAL.md -> live SKILL.md and marks the proposal applied.

Actual

Times out at the tool layer every time. After each attempt: proposal status: pending, live SKILL.md unchanged (no partial write). create/update (which write a similarly-sized PROPOSAL.md) work fine — only apply times out, which suggests the timeout is on apply's write/transition step, not on accepting the body.

Impact

Low severity (a clean manual workaround exists: verify still-pending, copy PROPOSAL.md to the live SKILL.md, reconcile proposals.json/proposal.json to applied). But to anyone who doesn't know to verify, it looks like 'apply failed' and may get retried or abandoned. Larger skill docs will keep hitting this.

Suggested fix

Raise the apply timeout, or stream/chunk the apply write, or make apply async with a status the caller can poll. A ~28KB doc write shouldn't time out.