Add tool_result pruning utility and split NL/T workflow into two passes #70

dsarno · 2025-09-07T00:42:22Z

Summary

add prune_tool_results.py to condense tool_result JSON outputs into one-line summaries
trim default Unity MCP tool payloads: find_in_file returns first match positions only, validate_script & get_sha provide condensed counts and hashes, read_console trims entries, and read_resource returns only hash and byte length unless text is requested
document lean tool responses and add tests for trimmed outputs
split NL/T GitHub workflow into separate NL and T passes to keep conversations lean

Testing

pytest

https://chatgpt.com/codex/tasks/task_e_68bcbbf232cc832791102f8596bd5683

coderabbitai · 2025-09-07T00:42:29Z

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

✨ Finishing Touches

🧪 Generate unit tests

Create PR with unit tests
Post copyable unit tests in a comment
Commit unit tests in branch codex/add-tool_result-json-pruner-gy2foz

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

greptile-apps

Greptile Summary

This PR implements a comprehensive optimization strategy for the Unity MCP (Multi-Client Protocol) system focused on reducing tool output verbosity and improving conversation efficiency. The changes introduce a tool result pruning utility (prune_tool_results.py) that condenses large JSON tool outputs into concise one-line summaries, making conversation logs more manageable and reducing token usage in AI interactions.

The core changes modify several Unity MCP tools to return "lean" responses by default: find_in_file now returns only match positions instead of full text excerpts and limits results to 1 match, validate_script provides warning/error counts rather than detailed diagnostics, get_sha returns only SHA256 hash and byte length, read_console defaults to error-only messages with trimmed fields, and read_resource returns metadata-only responses unless text content is explicitly requested. These modifications maintain full functionality through optional parameters while encouraging more efficient usage patterns.

Additionally, the GitHub CI workflow has been restructured to split the Claude NL/T test suite into two separate passes: an NL (Natural Language) pass using Claude 3.5 Sonnet for read-only operations, and a T (Transform) pass using Claude 3.5 Haiku for editing operations. This separation allows for more focused tool usage and keeps individual conversation sessions smaller and more manageable. The changes are well-documented in the README-DEV.md file and include comprehensive test coverage for the new trimmed behaviors.

Changed Files

Filename	Score	Overview
`.github/workflows/claude-nl-suite.yml`	4/5	Split workflow into separate NL and T passes using different Claude models and tool allowlists
`UnityMcpBridge/UnityMcpServer~/src/tools/resource_tools.py`	4/5	Modified read_resource and find_in_file to return minimal metadata by default with optional full content
`UnityMcpBridge/UnityMcpServer~/src/tools/read_console.py`	4/5	Changed defaults to return only error messages in JSON format with trimmed field output
`README-DEV.md`	5/5	Added comprehensive documentation for pruning utility, lean tool responses, and workflow changes
`tests/test_read_resource_minimal.py`	4/5	New test validating read_resource returns only SHA256 and byte length metadata by default
`UnityMcpBridge/UnityMcpServer~/src/tools/manage_script.py`	5/5	Modified validate_script and get_sha to return condensed summaries instead of full data
`tests/test_find_in_file_minimal.py`	4/5	New test ensuring find_in_file returns only position coordinates for matches
`tests/test_read_console_trimmed.py`	4/5	New test verifying read_console removes time fields and returns essential data only
`tests/test_validate_script_summary.py`	4/5	New test validating validate_script returns warning/error counts instead of full diagnostics
`tests/test_get_sha.py`	4/5	Updated existing test to verify get_sha returns only SHA256 and byte length fields
`prune_tool_results.py`	4/5	New utility script that condenses tool_result JSON outputs into one-line summaries for conversation management

Confidence score: 4/5

This PR is safe to merge with minimal risk as it focuses on output optimization without changing core functionality
Score reflects well-structured changes with comprehensive test coverage and clear documentation of the optimization strategy
Pay attention to the workflow file changes and ensure the split NL/T execution model works as expected in CI

Sequence Diagram

sequenceDiagram
    participant User
    participant GitHub as "GitHub Actions"
    participant Unity as "Unity Container"
    participant MCP as "MCP Server"
    participant Tools as "Unity Tools"
    
    User->>GitHub: "Trigger workflow dispatch"
    GitHub->>GitHub: "Setup Python/uv environment"
    GitHub->>Unity: "Start Unity container with MCP bridge"
    Unity->>MCP: "Initialize MCP connection on port"
    GitHub->>GitHub: "Write MCP config (.claude/mcp.json)"
    
    Note over GitHub: NL Pass
    GitHub->>MCP: "Run Claude NL pass (read-only tools)"
    MCP->>Tools: "find_in_file (returns positions only)"
    Tools-->>MCP: "{startLine, startCol, endLine, endCol}"
    MCP->>Tools: "validate_script (returns counts)"
    Tools-->>MCP: "{warnings: N, errors: N}"
    MCP->>Tools: "get_sha (returns hash + length)"
    Tools-->>MCP: "{sha256, lengthBytes}"
    MCP->>Tools: "read_console (trimmed entries)"
    Tools-->>MCP: "[{level, message}]"
    MCP-->>GitHub: "NL results (minimal payloads)"
    
    Note over GitHub: T Pass
    GitHub->>MCP: "Run Claude T pass (editing tools)"
    MCP->>Tools: "apply_text_edits"
    Tools-->>MCP: "Edit results"
    MCP->>Tools: "script_apply_edits"
    Tools-->>MCP: "Script modification results"
    MCP-->>GitHub: "T results"
    
    GitHub->>GitHub: "Canonicalize testcase names (NL/T prefixes)"
    GitHub->>GitHub: "Backfill missing tests as failures"
    GitHub->>GitHub: "Merge fragments into JUnit XML"
    GitHub->>GitHub: "Build markdown summary"
    GitHub->>GitHub: "Publish JUnit report and artifacts"
    GitHub->>Unity: "Stop Unity container"

_{11 files reviewed, 5 comments}

_{Edit Code Review Bot Settings | Greptile}

greptile-apps · 2025-09-07T00:43:12Z

tests/test_read_resource_minimal.py

+    loop = asyncio.new_event_loop()
+    try:
+        resp = loop.run_until_complete(
+            read_resource(uri="unity://path/Assets/A.txt", ctx=None, project_root=str(proj))
+        )
+    finally:
+        loop.close()


style: Consider using asyncio.run() instead of manual event loop management for cleaner async test execution

Suggested change

loop = asyncio.new_event_loop()

try:

resp = loop.run_until_complete(

read_resource(uri="unity://path/Assets/A.txt", ctx=None, project_root=str(proj))

)

finally:

loop.close()

resp = asyncio.run(

read_resource(uri="unity://path/Assets/A.txt", ctx=None, project_root=str(proj))

)

greptile-apps · 2025-09-07T00:43:20Z

tests/test_find_in_file_minimal.py

+    loop = asyncio.new_event_loop()
+    try:
+        resp = loop.run_until_complete(
+            find_in_file(uri="unity://path/Assets/A.txt", pattern="world", ctx=None, project_root=str(proj))
+        )
+    finally:
+        loop.close()


style: Consider using asyncio.run() instead of manually managing event loop lifecycle for cleaner code

greptile-apps · 2025-09-07T00:43:41Z

tests/test_validate_script_summary.py

+                    {"severity": "error"},
+                    {"severity": "fatal"},


logic: The test expects 'fatal' severity to be counted as an error, but this behavior should be verified against the actual implementation

greptile-apps · 2025-09-07T00:43:51Z

prune_tool_results.py

+    data = obj.get("data", {}) or {}
+    msg  = obj.get("message") or obj.get("status") or ""
+    # Common tool shapes
+    if "sha256" in str(data):


style: Using str(data) for SHA256 detection could match unrelated strings containing 'sha256'. Consider checking data.get('sha256') directly.

Suggested change

if "sha256" in str(data):

if data.get("sha256"):

greptile-apps · 2025-09-07T00:43:51Z

prune_tool_results.py

+        lines = data["lines"] or []
+        lvls = {"info":0,"warning":0,"error":0}
+        for L in lines:
+            lvls[L.get("level","" ).lower()] = lvls.get(L.get("level","" ).lower(),0)+1


style: Dictionary lookup pattern lvls.get(L.get('level','').lower(),0) is inefficient. Consider using defaultdict(int) or .setdefault().

Split NL/T workflow into two passes

28fcf01

dsarno added the codex label Sep 7, 2025 — with ChatGPT Codex Connector

greptile-apps bot reviewed Sep 7, 2025

View reviewed changes

dsarno closed this Sep 7, 2025

dsarno deleted the codex/add-tool_result-json-pruner-gy2foz branch September 7, 2025 03:22

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Add tool_result pruning utility and split NL/T workflow into two passes #70

Add tool_result pruning utility and split NL/T workflow into two passes #70

Uh oh!

dsarno commented Sep 7, 2025

Uh oh!

coderabbitai bot commented Sep 7, 2025

Review skipped

Uh oh!

greptile-apps bot left a comment

Uh oh!

greptile-apps bot Sep 7, 2025

Uh oh!

greptile-apps bot Sep 7, 2025

Uh oh!

greptile-apps bot Sep 7, 2025

Uh oh!

greptile-apps bot Sep 7, 2025

Uh oh!

greptile-apps bot Sep 7, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Add tool_result pruning utility and split NL/T workflow into two passes #70

Add tool_result pruning utility and split NL/T workflow into two passes #70

Uh oh!

Conversation

dsarno commented Sep 7, 2025

Summary

Testing

Uh oh!

coderabbitai bot commented Sep 7, 2025

Review skipped

Uh oh!

greptile-apps bot left a comment

Choose a reason for hiding this comment

Greptile Summary

Confidence score: 4/5

Sequence Diagram

Uh oh!

greptile-apps bot Sep 7, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Sep 7, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Sep 7, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Sep 7, 2025

Choose a reason for hiding this comment

Uh oh!

greptile-apps bot Sep 7, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants