m

Trace data lineage for a feature keyword from source to UI

Use this agent when a senior engineer has identified a bug or issue in the codebase and provided comments describing the problem. The agent will investigate the issue, propose a fix plan, wait for approval, and then implement the fix.\n\nExamples:\n\n<example>\nContext: A senior engineer has left comments about a race condition in the WebSocket handler.\nuser: "There's a race condition in the WebSocket message handler in back-end/app/services/ws_handler.py - when two clients send messages simultaneously, the session state gets corrupted because we're not locking around the state update. See lines 145-167."\nassistant: "I'll launch the bug-fix-engineer agent to investigate this race condition and propose a fix."\n<commentary>\nSince the user has described a specific bug with engineer-level detail, use the Task tool to launch the bug-fix-engineer agent to investigate, propose a fix plan, and implement it upon approval.\n</commentary>\n</example>\n\n<example>\nContext: A senior engineer has identified a data integrity issue.\nuser: "The renewal calculation in renewals/calculator.py is wrong - it's using the contract start date instead of the renewal date when computing the pro-rata amount. This causes incorrect invoices for mid-cycle renewals. The issue is in calculate_prorate() around line 89."\nassistant: "I'll use the bug-fix-engineer agent to analyze this calculation bug and propose a fix."\n<commentary>\nThe user has provided detailed engineer comments about a specific bug. Use the Task tool to launch the bug-fix-engineer agent to trace through the code, understand the issue, propose a fix, and implement it after approval.\n</commentary>\n</example>\n\n<example>\nContext: A senior engineer reports an issue found during code review.\nuser: "Found a bug in the frontend - the usePortfolioData hook in klair-client/src/hooks/usePortfolioData.ts has a stale closure issue. The filter callback on line 52 captures the old filterState but doesn't include it in the useCallback dependency array. This means filters don't actually apply until the user triggers a re-render."\nassistant: "I'll launch the bug-fix-engineer agent to investigate this stale closure issue and propose a fix."\n<commentary>\nSince the user is reporting a specific frontend bug with detailed engineer analysis, use the Task tool to launch the bug-fix-engineer agent to examine the hook, understand the closure issue, and propose a targeted fix.\n</commentary>\n</example>

Use this agent when you need to narrow down the files and code snippets involved in a particular bug or behavior, especially when it stems from changes having been made during the current session. It will run a verbatim command, and filter up the important bits. `git diff --staged`, comparing branches (`git diff main`), or `git diff` for unstaged changes.

Use this agent when the user wants to compare two documents to understand their differences, similarities, and relationship. This includes identifying whether documents are derived from the same base, finding what one document covers that the other doesn't, spotting differences of opinion, and summarizing commonalities.\n\nExamples:\n\n- User: "Compare these two CLAUDE.md files and tell me what's different"\n Assistant: "I'll use the doc-comparator agent to analyze both documents and provide a detailed comparison."\n [Launches doc-comparator agent via Task tool with both file paths]\n\n- User: "I have two API design specs - can you figure out what the second one is missing compared to the first?"\n Assistant: "Let me launch the doc-comparator agent to do a thorough comparison of the two specs."\n [Launches doc-comparator agent via Task tool]\n\n- User: "These two architecture docs look similar but I'm not sure what changed. Can you check?"\n Assistant: "I'll use the doc-comparator agent to determine if these are derived from the same base and identify all the changes."\n [Launches doc-comparator agent via Task tool]\n\n- User: "I found two competing design proposals - help me understand where they agree and disagree"\n Assistant: "I'll launch the doc-comparator agent to analyze both proposals and surface agreements and disagreements."\n [Launches doc-comparator agent via Task tool]

Use this agent when the user wants to investigate potential bugs, issues, or code quality concerns across the codebase. This agent launches parallel exploration tasks to trace through files, functions, and patterns involved in a reported issue, then evaluates whether each concern is valid or invalid. It marks findings with ✅ for valid issues and ❌ for invalid/non-issues. If provided a file as input, the prompt will specify which lines contain the bug reports to evaluate — the bot must read the file at those lines to understand each concern, then investigate and use the Edit tool to mark up the file in-place with QC_BOT_COMMENTS: annotations. IMPORTANT: When launching this agent with a file path, your Task prompt MUST instruct it to edit the file in-place using the Edit tool — do NOT tell it to 'return' findings as text. Use this agent when the user provides a bug report, a list of concerns, a file with comments/annotations about potential issues, or asks to validate whether something is actually broken.\n\nExamples:\n\n<example>\nContext: The user reports a potential bug in the renewal processing logic.\nuser: "I think there's a bug in how we calculate renewal dates - it seems to skip February in leap years"\nassistant: "Let me launch the qc-bugs agent to investigate this potential bug across the codebase and determine if it's valid."\n<commentary>\nSince the user reported a potential bug, use the Task tool to launch the qc-bugs agent to investigate the issue, trace the relevant code paths, and evaluate whether the bug is real.\n</commentary>\n</example>\n\n<example>\nContext: The user provides a file with review comments marking potential issues.\nuser: "Can you check if the issues flagged in this code review are actually valid? Here's the file: services/billing.py"\nassistant: "I'll use the qc-bugs agent to investigate each flagged issue and mark them as valid or invalid directly in the comments."\n<commentary>\nSince the user wants to validate review comments in a specific file, use the Task tool to launch the qc-bugs agent. It will investigate each comment, update them in-place with ✅ or ❌ markers.\n</commentary>\n</example>\n\n<example>\nContext: The user has a list of suspected issues they want validated.\nuser: "Here are 5 things I think might be wrong with the auth flow: 1) tokens aren't refreshed, 2) CORS is misconfigured, 3) rate limiting is missing, 4) sessions leak, 5) passwords stored in plaintext"\nassistant: "I'll launch the qc-bugs agent to investigate all 5 concerns in parallel and give you a validated summary."\n<commentary>\nSince the user provided a list of potential issues, use the Task tool to launch the qc-bugs agent which will spawn parallel exploration tasks for each concern and produce a summary with ✅/❌ markers.\n</commentary>\n</example>\n\n<example>\nContext: The user asks about a runtime error they encountered.\nuser: "We're getting a KeyError in the dashboard aggregation - is this a real bug or just bad test data?"\nassistant: "Let me use the qc-bugs agent to trace through the aggregation code paths and determine the root cause."\n<commentary>\nSince the user is asking about a potential bug, use the Task tool to launch the qc-bugs agent to investigate the code paths, data flow, and determine validity.\n</commentary>\n</example>

Use this agent when you have received reviewer comments on a PR and want to independently validate whether each comment is technically correct, materially significant, and worth acting on before responding or making changes. This agent triages reviewer feedback to separate genuine issues from noise.\n\nExamples:\n\n- User: "I just got review comments on my PR #1030, can you check if they're valid?"\n Assistant: "Let me use the PR Comment QC agent to independently validate each reviewer comment against the actual diff."\n [Launches pr-comment-qc agent via Task tool to analyze the PR diff and reviewer comments]\n\n- User: "Someone left 15 comments on my PR. Can you tell me which ones actually matter?"\n Assistant: "I'll use the PR Comment QC agent to triage all 15 comments and categorize them by significance."\n [Launches pr-comment-qc agent via Task tool]\n\n- User: "I think some of these review comments are wrong. Can you verify?"\n Assistant: "Let me launch the PR Comment QC agent to independently verify each comment's technical correctness against the actual code changes."\n [Launches pr-comment-qc agent via Task tool]\n\n- Context: User has just received a code review with mixed quality feedback including nits, duplicates, and potentially incorrect observations.\n User: "Got review feedback on PR 945, please QC it"\n Assistant: "I'll use the PR Comment QC agent to validate each comment and recommend how to handle them."\n [Launches pr-comment-qc agent via Task tool]

Create Claude Code agents using production-tested patterns and techniques

Relentlessly interview about a plan until shared understanding

Git worktree management with Git Town sync. Create worktrees, sync branches, push, propose PRs. Start/stop frontend+backend dev servers per worktree.

Parse a PR_comments file and launch parallel m:issue-qc agents for each unresolved issue giving each agent the PR_comments filepath and line numbers relevant to the issue it will investigate

Parse QC-annotated PR comments and launch bugfix agents in phased batches grouped by file domain