[FEATURE]: Inline image attachments don't include trigger text for skill auto-activation

[divitkashyap]

Problem

When OpenCode shows inline image previews in the chat, it displays just the filename (e.g., filename.png) without any surrounding trigger words like "analyze", "screenshot", "describe", or "look at".

Skills like vision-analysis are configured to auto-activate when the message contains these trigger words. However, since inline images only show the filename without any such context words, the skill never auto-activates.

Without trigger words: Agent doesn't know what to do → says "API requires valid key" and gives up.
With trigger words adjacent to image: Skill auto-activates → agent uses correct tool → works.

Expected Behavior

Skills with text trigger patterns should be able to auto-activate even when the image is shown inline, because the skill's activation depends on text adjacent to the image in the message.

Example

User sends: "[inline image: screenshot.png]"

OpenCode displays: Just screenshot.png with no trigger words

Agent receives: No trigger words in the text portion → vision-analysis skill doesn't activate

Proposed Solution

The inline image rendering should either:

1. Include trigger words in the text that gets sent to the agent (e.g., "look at this image: screenshot.png")
2. Extract trigger words from the filename itself (e.g., "Screenshot" from "Screenshot 2026-04-08 at 2.54.51 pm.png" normalizes to "screenshot" which IS a trigger word)
3. Or ensure that the skill activation logic considers image attachments as implicit triggers

This is an OpenCode-side limitation, not a skill configuration issue.

Investigation Summary

File to change: packages/opencode/src/session/message-v2.ts lines 654-669

Currently when stripMedia is true, it adds [Attached ${mime}: ${filename}] text, but for actual image input to the LLM, no trigger text is added before the file part.

The fix would add a text part with appropriate trigger words (extracted from filename or a fallback) before the file part, so skills like vision-analysis can auto-activate.

Verification

• This is a feature request for the core opencode application
• I have searched the existing issues and confirmed this is not a duplicate
• I understand this is a feature request and not a bug report

[github-actions]

This issue doesn't fully meet our contributing guidelines.

What needs to be fixed:

• No issue template was used. This project requires one of three templates: Bug Report, Feature Request, or Question. This issue appears to be a feature request or bug report but doesn't follow either template format.
• If this is a Bug Report, please use that template and include: Description, Steps to Reproduce, OpenCode version, Operating System, and Terminal fields.
• If this is a Feature Request, the title must start with [FEATURE]: and the template's verification checkbox must be included.

Please edit this issue to address the above within 2 hours, or it will be automatically closed.

If you believe this was flagged incorrectly, please let a maintainer know.

[github-actions]

This issue has been automatically closed because it was not updated to meet our contributing guidelines within the 2-hour window.

Feel free to open a new issue that follows our issue templates.

[divitkashyap]

Reopening with additional information after investigation.

Investigation Findings

After looking at the codebase, the fix location is clear:

File: packages/opencode/src/session/message-v2.ts lines 654-669

Currently when stripMedia is true, it adds [Attached ${mime}: ${filename}] text, but for actual image input to the LLM, no trigger text is added.

Proposed smarter fix: Extract trigger words from the filename itself:

• Screenshot 2026-04-08 at 2.54.51 pm.png → extract "Screenshot" → normalize to "screenshot" which IS a trigger word
• clipboard-2026-04-05.png → "clipboard" IS a trigger word
• Generic fallback: "screenshot" or "image" for files without obvious trigger words

This would be low-to-medium invasive change. The skill triggers are already designed to match these words - we just need to ensure they appear in the text sent to the LLM.

Verification

• This is a feature request for the core opencode application
• I have searched the existing issues and confirmed this is not a duplicate (this IS issue [FEATURE]: Inline image attachments don't include trigger text for skill auto-activation #21514 being updated)
• I understand this is a feature request and not a bug report

[divitkashyap]

I've updated this issue to follow the feature request template with the [FEATURE]: prefix and verification checkbox.

The investigation is complete - the fix location is identified at packages/opencode/src/session/message-v2.ts lines 654-669. This is a low-to-medium invasive change.

Would it be possible to reconsider this as "planned" or "open"? The issue describes a real limitation where screenshots from macOS (which naturally have "Screenshot" in the filename) don't trigger the vision-analysis skill because the word isn't passed to the LLM.

A simple fix would extract "screenshot" from filenames like "Screenshot 2026-04-08 at 2.54.51 pm.png" and include it in the text sent to the LLM.