Milens plan

milens-plan — Implementation Planning

Research the codebase end-to-end before writing code: understand structure, trace execution, assess blast radius, plan tests, and produce a structured plan following the ECC Planner specification.

Tools Required

Tool	Purpose
mcp_milens_codebase_summary	High-level project overview (languages, file count, domain count)
mcp_milens_domains	Module clusters showing logical groupings
mcp_milens_routes	API endpoints inventory with handler mappings
mcp_milens_smart_context	Intent-aware symbol analysis (edit intent)
mcp_milens_edit_check	Pre-edit safety: callers, export status, re-export chains, warnings
mcp_milens_trace	Execution flow from entrypoints to target
mcp_milens_impact	Blast radius: what breaks if target changes
mcp_milens_test_plan	Mock strategy + test scenarios for a symbol
mcp_milens_test_coverage_gaps	Related untested symbols

CRITICAL: All milens MCP tool calls MUST include the repo parameter set to the absolute path of the workspace root.

Workflow

Step 1: Project Overview

Understand the codebase at a high level.

mcp_milens_codebase_summary({repo: "<workspaceRoot>"})

Captures: languages, file counts, total symbols, top-level domains, entry points.

Step 2: Domain Structure

Map module clusters to understand logical groupings.

mcp_milens_domains({repo: "<workspaceRoot>"})

Shows which files form cohesive modules based on the dependency graph. Helps identify:

• Where the target code lives
• Which domains interact with it
• Natural boundaries for the change

Step 3: API Surface (optional)

If the project is a web service, inventory the endpoints.

mcp_milens_routes({repo: "<workspaceRoot>"})

Useful when the change touches:

• Route handlers
• Middleware
• Request/response types
• Authentication/authorization

Step 4: Target Analysis

Zoom in on the specific symbol to change.

mcp_milens_smart_context({name: "<symbolName>", intent: "edit", repo: "<workspaceRoot>"})

The edit intent returns: callers, export status, and immediate blast radius — only what matters for editing.

Step 5: Pre-Edit Safety Check

Run edit_check for pre-edit warnings.

mcp_milens_edit_check({name: "<symbolName>", repo: "<workspaceRoot>"})

This returns:

• Callers — who calls this symbol
• Export status — whether it's public API
• Re-export chains — barrel file re-exports that extend blast radius
• Warnings — breaking change risks, high fan-in, etc.

Step 6: Execution Flow

Trace how the target gets called at runtime.

mcp_milens_trace({name: "<symbolName>", direction: "to", repo: "<workspaceRoot>"})

Shows the call chains from entrypoints down to this symbol. Critical for understanding:

• What triggers this code
• What data flows into it
• What side effects it may have

Step 7: Blast Radius

Assess the full impact of the proposed change.

mcp_milens_impact({target: "<symbolName>", depth: 3, repo: "<workspaceRoot>"})

Shows all symbols that break if the target changes, up to the specified depth.

• If depth-1 dependents > 5: STOP and warn before proceeding
• If depth-3 dependents > 20: consider a facade or adapter pattern

Step 8: Test Strategy

Plan the testing approach.

mcp_milens_test_plan({name: "<symbolName>", repo: "<workspaceRoot>"})

Covers:

• Mock strategy (what to isolate)
• Test scenarios (happy path, edge cases, error handling)
• Affected test files

Step 9: Coverage Gaps (optional)

Check for related symbols that lack test coverage.

mcp_milens_test_coverage_gaps({repo: "<workspaceRoot>", limit: 10})

If neighboring symbols are untested, consider adding tests as part of the change.

Step 10: Produce Implementation Plan

Consolidate all findings into a structured plan:

1. Executive Summary — what changes, why, at what risk level
2. Codebase Context — domains, routes, and structure (from Steps 1-3)
3. Symbol Analysis — target details, callers, execution flow (from Steps 4-6)
4. Impact Assessment — blast radius, breaking change risks, dependents map (from Step 7)
5. Test Strategy — test plan, mock strategy, coverage gaps (from Steps 8-9)
6. Implementation Steps — ordered, actionable steps with file paths
7. Risk Mitigation — rollback plan, feature flags, phased deployment options

Example Session

Input

"plan a change to add rate limiting to the login handler"

Tool Calls

Step 1:

mcp_milens_codebase_summary({repo: "/home/user/project"})

Output: TypeScript project, 120 files, 850 symbols, 5 domains (auth, routes, db, utils, config).

Step 2:

mcp_milens_domains({repo: "/home/user/project"})

Output: auth domain contains login handler, token manager, session store.

Step 4:

mcp_milens_smart_context({name: "loginHandler", intent: "edit", repo: "/home/user/project"})

Output: 3 callers (Express route, rate limiter middleware, integration test), no re-exports.

Step 5:

mcp_milens_edit_check({name: "loginHandler", repo: "/home/user/project"})

Output: 3 callers, exported, no re-export chains, WARNING: exported function — changing signature is a breaking change.

Step 6:

mcp_milens_trace({name: "loginHandler", direction: "to", repo: "/home/user/project"})

Output: Chain: Express router → authMiddleware → loginHandler → tokenService.sign()

Step 7:

mcp_milens_impact({target: "loginHandler", depth: 3, repo: "/home/user/project"})

Output: Depth-1: 3 dependents (route, integration test, auth docs). Depth-2: 8 dependents. Depth-3: 14 dependents. No STOP threshold breached.

Step 8:

mcp_milens_test_plan({name: "loginHandler", repo: "/home/user/project"})

Output: Mock tokenService, userStore. 4 test scenarios: valid login, invalid password, rate limit exceeded, database failure.

Step 10 — Plan produced (see plan format above).

Best Practices

1. Never skip impact analysis. Even a "small" change can ripple through re-export chains. impact catches what intuition misses.
2. Use edit intent for smart_context when planning edits — it omits irrelevant detail and focuses on callers and blast radius.
3. Stop at depth-3 for impact. Deeper assessment rarely changes the implementation strategy but adds noise.
4. Trace before planning. Understanding the execution path prevents "add rate limiter after JWT decode" mistakes where ordering matters.
5. Write the plan before the code. The structured plan from Step 10 is the deliverable. Implementation follows the plan — not the other way around.

Quality Gate

Criteria	Pass	Fail
Codebase overview	All 3 overview tools succeed (summary, domains, routes)	Any tool fails or returns empty
Target analysis	smart_context + edit_check both run	Either tool skipped for the target symbol
Impact assessment	impact at depth 3 completed, dependents counted	Impact not run or depth < 3
Blast radius check	< 5 depth-1 dependents OR user warned and confirmed	> 5 depth-1 dependents without user confirmation
Test plan	test_plan covers happy path, edge cases, and errors	Less than 3 scenarios in the plan
Implementation plan	All 7 sections of the plan filled	Missing sections or incomplete analysis