DevAgentRadar → OpenCode

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

Context

Assistant: OpenCode Proposal: Resume failed OpenCode subagent work from the surfaced task_id Summary: OpenCode v1.18.20 surfaces failed subagent tool calls with a resumable task_id, answers subagent permission prompts in opencode run, and retries more network/xAI errors—so stop discarding empty failures and resume the same task instead. Primary source: https://github.com/anomalyco/opencode/releases/tag/v1.18.20

Why it matters

OpenCode v1.18.20 changes the subagent failure contract: failed tool calls now surface a resumable task_id instead of an empty result. Callers that treated empty output as a no-op will silently drop recoverable work. The same release answers permission requests triggered by subagents during opencode run, retries provider responses with finish_reason: network_error plus network-error/network_error variants, and retries xAI capacity and temporary unavailability stream errors. Cerebras max_completion_tokens is preserved without an extra output cap, so stacked limiters can truncate completions the provider already bounded.

Suggested actions

1. Upgrade and pin the OpenCode CLI to exactly v1.18.20 from https://github.com/anomalyco/opencode/releases/tag/v1.18.20 (OpenCode / SST).
2. In every OpenCode subagent tool-call handler, if the call fails, read the surfaced resumable task_id and resume that same task_id; do not spawn a replacement task and do not treat an empty result as success.
3. For unattended opencode run jobs that launch subagents, keep the run attached so permission requests triggered by those subagents are answered in-process instead of aborting or hanging the session.
4. On Cerebras-backed OpenCode sessions, set max_completion_tokens as the only output limit and delete any extra application-level output cap stacked on top of it.
5. Remove custom retry loops for finish_reason: network_error, the network-error/network_error variants, and xAI capacity or temporary-unavailability stream errors; v1.18.20 already retries those.

Config surfaces this release may change

• subagent definitions — check your repo before applying
• permission rules — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/opencode-v1-18-20-resume-failed-opencode-subagent-work-from-the-surfaced and mark Applied, Skipped, or Failed. Proposal id: faedb53a-a20a-43ef-b062-8e004a856ba3

Your job

1. Restate the change in one sentence.
2. Propose a minimal plan for my repo (or a throwaway pilot).
3. Implement only what I approve; prefer small diffs and tests.
4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.