Skip to content
Proposals/Upgrade OpenCode to v1.18.17 and wire Copilot PD
proposaleveryoneP5Worth a lookOpenCode

Upgrade OpenCode to v1.18.17 and wire Copilot PDF plus stable long sessions

Ship OpenCode v1.18.17 so session compaction keeps full recent turns, retries stop storming, Copilot PDFs attach when supported, and DeepSeek V4 Flash plus Muse models get correct defaults and prompts.

Why this loop

v1.18.17 fixes several workflow-breaking model and session behaviors: compaction now preserves complete recent turns and clearer summaries for smaller models; automatic retries are capped with jitter; GitHub Copilot models that advertise PDF vision can take PDF attachments; DeepSeek V4 Flash gets correct sampling defaults; Muse family models route to the Meta system prompt; MERGE Gateway reasoning variants work; desktop uses the server default model instead of stale local config. Teams should upgrade and explicitly point agent runs at these paths (PDF-backed Copilot reviews, long compacted sessions on smaller models, Muse/DeepSeek provider configs) rather than keeping pre-1.18.17 habits that hit retry storms, dropped context, or wrong prompts.

Proposed actions

  1. Upgrade OpenCode to v1.18.17 (CLI/desktop) and restart so session compaction, retry caps/jitter, Copilot PDF detection, DeepSeek V4 Flash sampling defaults, Muse→Meta system prompt routing, and MERGE Gateway reasoning variants all load.
  2. For document-heavy agents on GitHub Copilot models that advertise PDF vision, attach PDFs directly in the OpenCode session instead of pasting extracted text.
  3. For long threads on smaller models, continue in the same session and rely on v1.18.17 compaction (complete recent turns + clearer summaries) rather than manually trimming history mid-task.
  4. If you use DeepSeek V4 Flash or Muse-family models, select those provider model IDs after upgrade and do not override sampling or system prompt—OpenCode now applies correct Flash defaults and Meta prompts for all Muse variants.
  5. On desktop, clear or stop pinning a stale local default model so new sessions use the server’s current default model.

Agent prompt

Paste into your agent or query via MCP (`get_agent_prompt`) — free, no extra AI cost

Paste into Claude Code / CLAUDE.md task

# DevAgentRadar → Claude Code

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

## Context
Assistant: OpenCode
Proposal: Upgrade OpenCode to v1.18.17 and wire Copilot PDF plus stable long sessions
Summary: Ship OpenCode v1.18.17 so session compaction keeps full recent turns, retries stop storming, Copilot PDFs attach when supported, and DeepSeek V4 Flash plus Muse models get correct defaults and prompts.
Primary source: https://github.com/anomalyco/opencode/releases/tag/v1.18.17

## Why it matters
v1.18.17 fixes several workflow-breaking model and session behaviors: compaction now preserves complete recent turns and clearer summaries for smaller models; automatic retries are capped with jitter; GitHub Copilot models that advertise PDF vision can take PDF attachments; DeepSeek V4 Flash gets correct sampling defaults; Muse family models route to the Meta system prompt; MERGE Gateway reasoning variants work; desktop uses the server default model instead of stale local config. Teams should upgrade and explicitly point agent runs at these paths (PDF-backed Copilot reviews, long compacted sessions on smaller models, Muse/DeepSeek provider configs) rather than keeping pre-1.18.17 habits that hit retry storms, dropped context, or wrong prompts.

## Suggested actions
1. Upgrade OpenCode to v1.18.17 (CLI/desktop) and restart so session compaction, retry caps/jitter, Copilot PDF detection, DeepSeek V4 Flash sampling defaults, Muse→Meta system prompt routing, and MERGE Gateway reasoning variants all load.
2. For document-heavy agents on GitHub Copilot models that advertise PDF vision, attach PDFs directly in the OpenCode session instead of pasting extracted text.
3. For long threads on smaller models, continue in the same session and rely on v1.18.17 compaction (complete recent turns + clearer summaries) rather than manually trimming history mid-task.
4. If you use DeepSeek V4 Flash or Muse-family models, select those provider model IDs after upgrade and do not override sampling or system prompt—OpenCode now applies correct Flash defaults and Meta prompts for all Muse variants.
5. On desktop, clear or stop pinning a stale local default model so new sessions use the server’s current default model.


## 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.

Your loop

This browser · no sign-in · not shared as “you”

Your decision stays on this device. A public tally appears after a few votes.

Originating release signal

OpenCodev1.18.17Aug 12, 2026

v1.18.17

Made session compaction keep complete recent turns and produce clearer summaries for smaller models. · Added MERGE Gateway reasoning variants so those model options work correctly. (@MatthewFeroz) · Capped automatic session retries and added jitter to reduce repeated retry storms. · Enabled PDF attachments for GitHub Copilot models that advertise PDF vision support. (@stevenao) · Applied the correct sampling defaults to DeepSeek V4 Flash on supported providers. · Routed all Muse family models to the correct Meta system prompt. (@mreso) · +16 more changes

Verified excerpt — the source's own words
## Core

### Bugfixes

- Made session compaction keep complete recent turns and produce clearer summaries for smaller models.
- Added MERGE Gateway reasoning variants so those model options work correctly. (@MatthewFeroz)
- Capped automatic session retries and added jitter to reduce repeated retry storms.
- Enabled PDF attachments for GitHub Copilot models that advertise PDF vision support. (@stevenao)
- Applied the correct sampling defaults to DeepSeek V4 Flash on supported providers.
- Routed all Muse family models to the correct Meta system prompt. (@mreso)

## Desktop

### Bugfixes

- Updated Chinese developer terminology to use more widely recognized translations.
- Used the server's current default model instead of stale local config when choosing the default model.

**Thank you to 6 community contributors:**
- @mreso:
  - fix(session): route all Muse family models to the Meta system prompt (#41581)
- @OpeOginni:
  - fix(ui): correct OC-2 weak icon color (#41504)
- @peculiarnewbie:
  - feat(console): add go usage endpoint (#16513)
- @stevenao:
  - fix(opencode): detect Copilot PDF input support (#41522)
- @MatthewFeroz:

Excerpt ends here — this release continues at the source ↗.

Primary source ↗