using-leankg

name	using-leankg
description	Use for ANY code search, navigation, or finding code logic - "where is X", "find logic Y", "how does Z work", impact analysis, dependencies. LeanKG is MANDATORY first.

<EXTREMELY_IMPORTANT> STRICT ENFORCEMENT: You MUST follow the 4-Tier Search Fallback Chain EXACTLY.

1. ALWAYS start with mcp_status to check LeanKG readiness
2. If NOT ready, invoke mcp_init or mcp_index to initialize
3. Use LeanKG MCP tools for ALL searches (Tier 1)
4. ONLY fall back to leankg CLI if MCP fails (Tier 2)
5. ONLY fall back to rtk if CLI fails (Tier 3)
6. ONLY fall back to grep/rg as ABSOLUTE LAST RESORT (Tier 4)

NO EXCEPTIONS. NO RATIONALIZATION. Follow the chain. </EXTREMELY_IMPORTANT>

MANDATORY 4-Tier Search Fallback Chain

When user asks "where is X", "find logic Y", "how does Z work", etc:
           |
           v
TIER 1: LeanKG MCP Server (leankg_mcp_* tools)
           |
           mcp_status (ALWAYS check first)
           search_code("X") or find_function("X") or query_file("X")
           |
           +-- Results found --> get_context(file) to read content --> DONE
           +-- MCP down/error --> TIER 2
           |
           v
TIER 2: leankg CLI command
           |
           leankg query "X" --kind name
           |
           +-- Results found --> DONE
           +-- Empty/error --> TIER 3
           |
           v
TIER 3: rtk (rich toolkit grep)
           |
           rtk grep "X" --path .
           rtk file "pattern" --path .
           |
           +-- Results found --> DONE
           +-- Empty --> TIER 4
           |
           v
TIER 4: grep/rg (ABSOLUTE LAST RESORT)
           |
           rg "X"
           grep -rn "X" --include="*.ext"

ABSOLUTE BANS

• NEVER skip directly to grep/rg when LeanKG is available
• NEVER use Glob to find code files before checking LeanKG
• NEVER use Grep tool before exhausting Tiers 1-3
• The ONLY exception: searching non-code files (config, docs, data files)

Tier 1: LeanKG MCP Tools (Use in this order)

Step	Tool	When to Use
1	mcp_status	ALWAYS check first
2	search_code("X")	Find code by name/type
3	find_function("X")	Locate function definitions
4	query_file("*X*")	Find files by name
5	get_impact_radius(file)	Blast radius for changes
6	get_context(file)	READ file content (token-optimized)
7	get_dependencies(file)	Get imports
8	get_dependents(file)	Get reverse dependencies
9	get_tested_by(file)	Find tests
10	get_callers("func")	Find who calls a function
11	get_call_graph("func")	Full call graph

Tier 2: leankg CLI (Only if MCP fails)

leankg status
leankg query "X" --kind name
leankg impact file 3

Tier 3: rtk Fallback (Only if leankg CLI empty)

rtk grep "X" --path .
rtk file "pattern" --path .

Tier 4: grep/rg (ABSOLUTE LAST RESORT)

rg "X"
grep -rn "X" --include="*.ext"

Critical: After search_code returns file paths

IMPORTANT: When search_code returns results with file paths:

1. Use get_context(file_path) to READ the actual file content
2. Do NOT just report the file paths - show the code

Common Triggers

User says...	Start with
"where is X"	search_code("X") or find_function("X")
"find the logic"	search_code("logic_name")
"how does X work"	get_context(file) after search_code
"what calls X"	get_callers("X") or get_call_graph("X")
"what breaks if I change X"	get_impact_radius("X")
"find all files named X"	query_file("X")
"who imports X"	get_dependents("X")
"what does X depend on"	get_dependencies("X")