feat: add file persistence for large tool outputs #6234
+112
−45
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Previously, large outputs from tools were either truncated (bash) or passed
through entirely to the model context (MCP, webfetch, websearch, codesearch),
causing data loss or context overflow issues respectively.
This change implements a file persistence pattern where outputs exceeding
30,000 characters are saved to disk and the model receives a reference
with instructions for exploring the data using Read, Grep, or jq tools.
Changes:
- Add tool-results.ts: Storage helper that saves large outputs as plain
text files to storage/tool_results/{sessionID}/{toolName}-{timestamp}.txt
- Update prompt.ts: MCP tools now save large outputs to file instead of
passing them directly to model context
- Update bash.ts: Replace simple truncation with file persistence,
preserving full output instead of losing data beyond 30k chars
- Update webfetch.ts: Add file persistence for large fetched content
- Update websearch.ts: Add file persistence for large search results
- Update codesearch.ts: Add file persistence for large code search results
- Update bash.txt: Update tool description to reflect new behavior
- Update session/index.ts: Clean up tool_results directory when session
is deleted using fs.rm with recursive:true
Benefits:
- No data loss: full output is preserved on disk
- Context stays small: only file path and instructions sent to model
- Model can explore large outputs using existing tools (Read, Grep, jq)
- Consistent with existing session data lifecycle (cleaned up on delete)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Implements file persistence for large tool outputs instead of truncating and losing data. When tool outputs exceed 30,000 characters, the full output is saved to disk and the model receives a file path with instructions for exploring the data using Read, Grep, or jq tools.
This addresses the root cause of "prompt is too long" errors caused by large tool outputs overwhelming the context window.
Closes #4560
Related: #4826, #4845, #5360
Changes
tool-results.ts: Storage helper that saves large outputs to~/.local/share/opencode/storage/tool_results/{sessionID}/{toolName}-{ulid}.txtprompt.ts: MCP tools now save large outputs to file instead of passing directly to model contextbash.ts: Replace simple truncation with file persistence, preserving full output instead of losing data beyond 30k chars. Added 10MB hard limit to prevent memory issues.webfetch.ts: Add file persistence for large fetched contentwebsearch.ts: Add file persistence for large search resultscodesearch.ts: Add file persistence for large code search resultsbash.txt: Tool description reflects new behaviorsession/index.ts: Clean up tool_results directory when session is deletedOPENCODE_EXPERIMENTAL_MCP_MAX_OUTPUT_LENGTHfor configurabilityBenefits
Coverage
This PR covers file persistence for: