bobbin

name	bobbin
description	Semantic code search and codebase intelligence. Use this skill when asked to "search code", "find where X is defined", "what files are related to Y", "show me the impact of changing Z", "find duplicates", "search commits", "search beads/issues", "/bobbin <query>", or any codebase exploration that goes beyond simple grep. Routes queries to the right bobbin MCP tool.

Bobbin Skill — Semantic Code Search

You are performing a codebase search using bobbin's semantic search engine. Bobbin indexes code across all repos in the workspace and provides tools far more powerful than grep: semantic search, symbol references, impact analysis, commit history, coupling graphs, and more.

Input

The user provided a query: {{args}}

Routing Decision

Analyze the query and pick the RIGHT tool. Don't default to search for everything — use the most specific tool available.

Route 1: Symbol Lookup

Trigger: Query names a specific function, struct, class, or variable. Examples: "find parse_config", "where is Config defined", "who calls handle_request"

Action: Use bobbin find_refs with the symbol name.

• Returns: definition location + all usage sites
• Follow up with bobbin read_chunk to show the actual code

Route 2: Semantic Code Search

Trigger: Query describes code by what it does, not by exact name. Examples: "authentication logic", "error handling with retries", "database connection pooling"

Action: Use bobbin search with the natural language query.

• Use mode: "hybrid" (default) for best results
• Use type filter if the user wants a specific kind (function, struct, etc.)
• Use repo filter if the user specifies a repo
• Use bundle filter if the user specifies a bundle scope

Route 3: Exact Pattern Search

Trigger: Query is a literal string, regex, or exact identifier. Examples: "TODO FIXME", "fn.*async", grep-style patterns

Action: Use bobbin grep with the pattern.

• Set regex: true if the pattern uses regex syntax
• Set ignore_case: true for case-insensitive searches

Route 4: File Relationships

Trigger: Query asks what files are related, coupled, or co-changed. Examples: "what changes with config.rs", "files related to auth"

Action: Use bobbin related with the file path.

• Follow up with bobbin read_chunk on interesting results

Route 5: Impact Analysis

Trigger: Query asks what would break or be affected by a change. Examples: "impact of changing auth.rs", "what breaks if I modify the parser"

Action: Use bobbin impact with the target file or file:function.

• Use depth: 2 for transitive impact

Route 6: Commit Search

Trigger: Query asks about git history semantically. Examples: "when was auth added", "commits that changed error handling", "who refactored X"

Action: Use bobbin commit_search with the query.

• Use author filter if specified
• Use file filter if specified

Route 7: Bead/Issue Search

Trigger: Query asks about issues, bugs, tasks, or work items. Examples: "open auth bugs", "P1 issues", "beads about monitoring"

Action: Use bobbin search_beads with the query.

• Use priority/status/assignee/label filters as appropriate

Route 8: Context Assembly

Trigger: Query asks for comprehensive context for a task or change. Examples: "context for refactoring auth", "everything related to the deploy pipeline"

Action: Use bobbin context with the task description.

• Adjust budget based on scope (default 500 lines)

Route 9: File Symbols

Trigger: Query asks what's defined in a specific file. Examples: "symbols in main.rs", "what functions are in config.py", "API of auth module"

Action: Use bobbin list_symbols with the file path.

Route 10: Duplicate Detection

Trigger: Query asks about duplicate or similar code. Examples: "find code similar to X", "detect duplicates", "near-duplicate functions"

Action: Use bobbin similar.

• Single target: provide target as file:function or free text
• Scan mode: set scan: true to find clusters across the codebase

Route 11: Dependency Analysis

Trigger: Query asks about imports, dependencies, or what depends on a file. Examples: "what imports config.rs", "dependencies of auth module", "reverse deps"

Action: Use bobbin dependencies with the file path.

• Use reverse: true for "what depends on this"
• Use both: true for full picture

Route 12: File History

Trigger: Query asks about the history of a specific file. Examples: "history of config.rs", "who changed auth.py recently", "churn on main.rs"

Action: Use bobbin file_history with the file path.

Route 13: Code Hotspots

Trigger: Query asks about risky, complex, or high-churn code. Examples: "hotspots", "what needs refactoring", "riskiest files"

Action: Use bobbin hotspots.

• Adjust since for time window
• Adjust threshold to filter noise

Route 14: Archive Search

Trigger: Query asks about chat logs, IRC discussions, or agent memory. Examples: "what did we discuss about auth", "recent deploy discussions", "telegram messages"

Action: Use bobbin archive_search with the query.

• Use source: "hla" for chat logs, source: "pensieve" for agent memory
• Use date filters (after, before) as appropriate

Route 15: Review Context

Trigger: Query asks for review context on current changes or a diff. Examples: "review context for my changes", "what do I need to review this PR"

Action: Use bobbin review.

• Use diff: "staged" or diff: "unstaged" for working tree
• Use diff: "branch:feature-x" for branch comparison

Output Guidelines

1. Show the most relevant results first — don't dump everything
2. Include file paths and line numbers so the user can navigate
3. Read and show actual code for the top 2-3 results using read_chunk
4. Suggest follow-up queries if the results are ambiguous or partial
5. Chain tools when appropriate — e.g., search -> find_refs -> read_chunk
6. Be concise — summarize what you found, don't just paste raw output