Problem

OpenCode CLI runtime (src/middleware/runtimes/opencode.ts) has a --file flag for file input, but it hardcodes text/plain MIME type for all files (source). Binary media (images, audio, video) is sent as text, making multimodal input non-functional.

Current state

• Uses run --format json for invocation
• --file <path> flag exists but hardcodes text/plain MIME type
• All binary media would be corrupted/ignored

Implementation

While upstream bug persists

• Declare: acceptsInbound: [] (no media support)
• All media must be handled by middleware fallback (STT for audio, vision API for images)
• emitsOutbound: false

readonly mediaCapabilities = {
  acceptsInbound: [],
  emitsOutbound: false,
};

When upstream fixes MIME detection

• Pass --file <path> for media attachments with correct MIME types
• Temp file management similar to Gemini
• Update capability based on which MIME types OpenCode supports

Tests

Unit tests (src/middleware/runtimes/opencode.test.ts)

• execute() with media present does NOT pass --file flag (upstream bug workaround)
• mediaCapabilities reports empty acceptsInbound

Future tests (when upstream is fixed)

• execute() with image attachment passes --file flag
• Live smoke test: send image, verify OpenCode describes content

Upstream tracking

• Filed upstream: bug: run --file hardcodes text/plain MIME type for all files anomalyco/opencode#16723
• When fixed: update mediaCapabilities, implement --file passthrough, enable tests

Depends on

• feat(middleware): add multimodal media attachments to AgentRuntime contract #385 — AgentRuntime multimodal contract

Related

• feat(runtimes): implement multimodal I/O for each CLI runtime #386 — per-runtime multimodal (parent tracking issue)
• research(auth+media): auth profiles and media understanding architecture #376 — auth/media restructuring