Interactive Mention Response

📋 Overview

Responds to @overcut mentions in issues, PRs, and comments with context-aware answers using multi-agent coordination. Automatically identifies relevant repositories, clones them for full code access, and opens an interactive session where a coordinator delegates questions to specialized agents — providing concise, evidence-based answers grounded in your actual codebase.

⚡ Triggers

Automatic:

• Event: mention — triggers whenever @overcut is mentioned in any issue, PR, or comment
• No conditions — responds to all mentions
• No delay

🎯 Use Cases

• Q&A: Ask questions about code, architecture, or implementation details
• Debugging: Investigate bugs, trace error paths, and identify root causes
• Code explanation: Understand what a file, function, or module does and why
• Impact analysis: Assess how a change affects the rest of the codebase
• Investigation: Trace data flows, find usages, and explore dependencies
• Code suggestions: Get concrete fix proposals or improvement recommendations
• PR context: Understand what a PR changes, why, and what it affects

🔧 Prerequisites

• Agents configured:
  • Product Manager — Understands requirements and business context
  • DevOps Engineer — Infrastructure, CI/CD, and deployment expertise
  • Senior Developer — Code analysis, architecture, and implementation
  • Code Reviewer — Code quality, patterns, and best practices
  • Root-Cause Analysis (RCA) Expert — Debugging, failure analysis, and incident investigation
  • Technical Writer — Clear documentation and structured communication

🏗️ Workflow Steps

1. Identify Repositories (repo.identify) — Finds relevant repos based on the mention context

   • Agents: None (automated repository identification)
   • Duration: ~30 seconds
   • Returns up to 3 repositories with minimum 0.4 confidence
   • Prioritizes the component field for identification

2. Clone Repo (git.clone) — Clones identified repositories

   • Agents: None (automated git operation)
   • Duration: ~1 min
   • Shallow clone (depth 1, single branch) for efficiency

3. Multi-Agent Session (agent.session) — Interactive session to answer the user's question

   • Agents: Product Manager, DevOps Engineer, Senior Developer, Code Reviewer, RCA Expert, Technical Writer (coordinated by Coordinator)
   • Duration: Up to 120 min (interactive session)
   • Process:
     1. Parse & Plan: Extract intent, scope, and artifacts from the @overcut mention
     2. Gather Evidence: Read diffs, search code, open relevant files
     3. Respond: Deliver concise, cited answer with supporting details
     4. Follow-up: Keep session open for continued conversation
   • Listens for follow-up comments in the same thread
   • Session remains open until /done, "thanks", or timeout

[Identify Repos] → [Clone Repo] → [Multi-Agent Session]
                                          ↕
                                   (listens for follow-up
                                    @overcut comments)

🔑 Key Features

• Interactive session: The session stays open and listens for follow-up comments, enabling a conversational flow without restarting the workflow
• Comment listening: Responds to subsequent @overcut mentions in the same thread with full prior context
• Multi-agent coordination: The coordinator delegates to the best-suited agent for each question (e.g., RCA Expert for debugging, Senior Developer for code analysis)
• Read-only by default: Agents browse code and analyze but do not push changes or modify settings unless explicitly asked
• Evidence-based answers: Every response cites specific files, lines, commits, or diffs

📝 Response Format

Each response follows a structured format:

Answer

• Concise result: facts, decision, or fix
• Minimal code snippets when helpful

Why this is correct

• Source: path/to/file.ext:LINE-START–LINE-END (brief rationale)
• PR/Commit references when relevant

Next steps (if applicable)

• Actionable checklist items

🎨 Customization

Step Prompt

• agent-session.md — Controls the coordinator's behavior, response format, operating rules, and failure handling

Agents

Swap or add agents in workflow.json under the agent-session step's agentIds array and in refs.agents. For example:

• Add a Database Architect for data-layer questions
• Add a Security Engineer for vulnerability-related mentions
• Remove agents you don't need to reduce coordination overhead

Exit Criteria

Edit the exitCriteria in workflow.json to adjust:

• maxDurationMinutes — Session timeout (default: 120 min)
• userSignals.explicit — Commands that end the session (default: /done, thanks)

Common Adjustments

Change response style: Edit agent-session.md Response Format section to:

• Add project-specific sections (e.g., "Performance Impact", "Security Considerations")
• Adjust verbosity level
• Change citation format

Restrict scope: Edit agent-session.md Operating Rules to:

• Limit to specific repos or directories
• Add domain-specific guidelines
• Enforce organizational policies

Add conditions to trigger: Edit the trigger in workflow.json to filter mentions:

• Only respond in specific repos
• Only respond to certain users or teams
• Require specific labels on the issue/PR

🔗 Related Workflows

• Code Review — Automated code review for PRs
• Auto Root Cause Analysis — Automated failure investigation triggered by CI failures
• Remediate CVEs — Security vulnerability analysis and remediation planning
• Auto PR Description — Automatically generates PR descriptions with context

---

Part of the Overcut Playbooks collection