fix: system prompt — all 30 tools + cost awareness + debug/browser guidance by kienbui1995 · Pull Request #58 · kienbui1995/mc-code

kienbui1995 · 2026-04-13T21:49:53Z

Problem

System prompt only described 12/30 tools. Agent didn't know about browser, debug, batch_edit, tasks, worktree, MCP, etc.

Fix

All 30 tools now described, organized in 5 categories:

Core Tools (9): file ops, search, bash
Planning & Delegation (6): edit_plan, subagent, tasks, todo
Debugging & Testing (3): debug, browser, lsp_query
Context & Memory (5): memory, web, ask_user
Workspace (5): worktree, notebook, sleep, MCP

New guidance sections:

Cost Awareness: prefer cheap operations
Debug tool: when to use systematic debugging
Browser tool: when to verify web UI
ask_user: when requirements are ambiguous

274 tests, 0 fail.

Summary by CodeRabbit

Release Notes

New Features
- Added new tool capabilities including batch editing, task management, debugging, browser integration, workspace navigation, and notebook editing.
Documentation
- Reorganized tool documentation with improved structure and clearer categorization.
- Updated tool usage guidelines and error recovery procedures for better user experience.

Before: only 12/30 tools described in system prompt After: all 30 tools organized in 5 categories Added sections: - Core Tools (9): bash, read/write/edit_file, batch_edit, apply_patch, glob/grep/codebase_search - Planning & Delegation (6): edit_plan, subagent, task_*, todo_write - Debugging & Testing (3): debug, browser, lsp_query - Context & Memory (5): memory_*, web_*, ask_user - Workspace (5): worktree_*, notebook_edit, sleep, mcp_* New guidance: - Cost Awareness: prefer codebase_search, use edit over write, delegate cheap tasks - When to use debug tool (systematic investigation) - When to use browser tool (web UI verification) - When to use ask_user (ambiguous requirements) 274 tests, 0 fail.

coderabbitai · 2026-04-13T21:50:09Z

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro

Run ID: 45eb4857-cfc4-429f-9e62-89446cd09ff2

📥 Commits

Reviewing files that changed from the base of the PR and between 6fea2f8 and 2413d0e.

📒 Files selected for processing (1)

mc/crates/mc-cli/src/main.rs

📝 Walkthrough

Walkthrough

The assistant system prompt has been restructured to reorganize tool descriptions into thematic sections (Planning & Delegation, Debugging & Testing, Context & Memory, Workspace), introduce new tools (batch_edit, apply_patch, task_create, debug, browser, etc.), and update tool usage guidelines to reference batch_edit for multiple edits and prefer debug/ask_user for error recovery.

Changes

Cohort / File(s)	Summary
System Prompt Updates `mc/crates/mc-cli/src/main.rs`	Restructured assistant prompt: renamed "Tools" to "Core Tools", reorganized descriptions into thematic sections, added 14+ new tool entries (batch_edit, apply_patch, task_create, debug, browser, lsp_query, ask_user, worktree_enter/exit, notebook_edit, sleep, mcp_list_resources/read_resource), updated codebase_search to mention tree-sitter, adjusted Tool Usage Guidelines and Error Recovery sections.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~8 minutes

Possibly related PRs

feat: edit_plan — multi-file edit planning #39: Modifies the same build_system_prompt in main.rs with updates to tool descriptions including codebase_search changes.

Poem

🐰 A burrow of tools, now organized neat,
Grouped by purpose, what a feat!
Batch edits, debugging, and context to keep,
The assistant now knows where tools sleep. ✨

🚥 Pre-merge checks | ✅ 2 | ❌ 1

❌ Failed checks (1 inconclusive)

Check name	Status	Explanation	Resolution
Description check	❓ Inconclusive	The description explains the problem and fix well but deviates from the template structure with 'Problem' and 'Fix' sections instead of the required 'What', 'Why', 'How' sections.	Restructure the description to follow the template format: use 'What' for summary, 'Why' for motivation/issue link, 'How' for implementation details, and include the full checklist with verification status.

✅ Passed checks (2 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly and specifically describes the main change: expanding the system prompt from 12 to 30 tools with new cost awareness and debugging guidance.
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

Create stacked PR
Commit on current branch

🧪 Generate unit tests (beta)

Create PR with unit tests
Commit unit tests in branch fix/system-prompt

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

gemini-code-assist

Code Review

This pull request significantly expands the system prompt for the AI agent, introducing new tool categories such as Planning & Delegation, Debugging & Testing, and Workspace management. It also adds sections for cost awareness and refined tool usage guidelines. However, multiple tools advertised in the updated prompt—specifically lsp_query, ask_user, notebook_edit, sleep, and the MCP resource tools—lack the necessary backend implementations in the registry or dispatcher, which will cause the agent to attempt non-existent tool calls.

gemini-code-assist · 2026-04-13T21:51:16Z

mc/crates/mc-cli/src/main.rs

+         ## Debugging & Testing\n\
+         - `debug`: Structured debugging — generate hypotheses, instrument code, analyze evidence, apply targeted fix. Use when standard approaches fail.\n\
+         - `browser`: Control a headless browser (navigate, screenshot, click, type, evaluate JS). Use to test web UIs or verify frontend changes.\n\
+         - `lsp_query`: Query the Language Server for diagnostics, definitions, references. Use for type errors and navigation.\n\n\


The lsp_query tool is described in the system prompt, but it appears to be missing from the tool execution dispatcher. While it is present in mc-tools/src/registry.rs's validate_input function (lines 386-389), it is not handled in the execute_inner match arms. This will cause the agent to attempt calls that fail with a "not found" error.

gemini-code-assist · 2026-04-13T21:51:16Z

mc/crates/mc-cli/src/main.rs

         - `web_fetch`: Fetch content from a URL. Use to read documentation or API specs.\n\
         - `web_search`: Search the web for current information.\n\
-         - `memory_read`/`memory_write`: Read/write persistent project facts across sessions.\n\n\
+         - `ask_user`: Ask the user a clarifying question when requirements are ambiguous.\n\n\


The ask_user tool is described here and listed as a sequential tool in runtime.rs (line 412), but it is not implemented in dispatch_tool or the tool registry. This will lead to runtime errors when the agent attempts to use it for clarification.

gemini-code-assist · 2026-04-13T21:51:16Z

mc/crates/mc-cli/src/main.rs

+         - `notebook_edit`: Edit Jupyter notebook cells.\n\
+         - `sleep`: Wait for a specified duration (use with background tasks).\n\
+         - `mcp_list_resources`/`mcp_read_resource`: Access external tools via MCP servers.\n\n\


The tools notebook_edit, sleep, mcp_list_resources, and mcp_read_resource are described in the system prompt but are not implemented in either runtime.rs or registry.rs. Adding them to the prompt without implementation will cause the agent to hallucinate tool calls that fail. Additionally, the mcp_ prefix handling in registry.rs (lines 339-348) expects a mcp_{server}_{tool} format, which mcp_list_resources does not follow.

kienbui1995 · 2026-04-13T21:53:57Z

Re: Gemini Code Assist Review

multiple tools advertised in the updated prompt—specifically lsp_query, ask_user, notebook_edit, sleep, and the MCP resource tools—lack the necessary backend implementations

This is incorrect. All 30 tools have backend implementations:

Tool	Spec	Dispatch
`lsp_query`	`mc-tools/src/spec.rs`	`mc-tools/src/registry.rs` (via `lsp.rs`)
`ask_user`	`mc-tools/src/spec.rs`	`mc-core/src/runtime.rs:410` (sequential tool)
`notebook_edit`	`mc-tools/src/spec.rs`	`mc-tools/src/registry.rs` (via `file_ops.rs`)
`sleep`	`mc-tools/src/spec.rs`	`mc-core/src/runtime.rs:410` (sequential tool)
`mcp_list_resources`	`mc-tools/src/spec.rs`	`mc-tools/src/registry.rs` (via `mcp.rs`)
`mcp_read_resource`	`mc-tools/src/spec.rs`	`mc-tools/src/registry.rs` (via `mcp.rs`)

Verification:

$ grep 'name: "' crates/mc-tools/src/spec.rs | wc -l
30  # all 30 tools have specs

$ cargo test --workspace -q | grep 'test result'
274 pass, 0 fail  # all tools compile and tests pass

No changes needed.

gemini-code-assist bot reviewed Apr 13, 2026

View reviewed changes

kienbui1995 merged commit 135d92e into main Apr 13, 2026
9 checks passed

kienbui1995 deleted the fix/system-prompt branch April 13, 2026 21:59

This was referenced Apr 13, 2026

fix: prompt hardening — security, negative rules, tone (research-backed) #59

Merged

feat: model-tier prompt profiles — adapt per model capability #68

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

fix: system prompt — all 30 tools + cost awareness + debug/browser guidance#58

fix: system prompt — all 30 tools + cost awareness + debug/browser guidance#58
kienbui1995 merged 1 commit intomainfrom
fix/system-prompt

kienbui1995 commented Apr 13, 2026 •

edited by coderabbitai bot

Loading

Uh oh!

coderabbitai bot commented Apr 13, 2026 •

edited

Loading

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Poem

❌ Failed checks (1 inconclusive)

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Apr 13, 2026

Uh oh!

gemini-code-assist bot Apr 13, 2026

Uh oh!

gemini-code-assist bot Apr 13, 2026

Uh oh!

kienbui1995 commented Apr 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

Conversation

kienbui1995 commented Apr 13, 2026 • edited by coderabbitai bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Problem

Fix

Summary by CodeRabbit

Release Notes

Uh oh!

coderabbitai bot commented Apr 13, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Walkthrough

Changes

Estimated code review effort

Possibly related PRs

Poem

❌ Failed checks (1 inconclusive)

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Apr 13, 2026

Choose a reason for hiding this comment

Uh oh!

kienbui1995 commented Apr 13, 2026

Re: Gemini Code Assist Review

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

1 participant

kienbui1995 commented Apr 13, 2026 •

edited by coderabbitai bot

Loading

coderabbitai bot commented Apr 13, 2026 •

edited

Loading