Milens code review

milens-code-review — Automated Code Review

Analyze a PR or local changes with multi-layered review: risk scoring, symbol deep-dives, dead code audit, tech debt search, and a structured report.

Tools Required

Tool	Purpose
mcp_milens_review_pr	List changed files + affected symbols with risk scores
mcp_milens_review_symbol	Deep-dive on a single symbol (role, heat, dependents, test status)
mcp_milens_context	Incoming references + outgoing dependencies
mcp_milens_find_dead_code	Find exported symbols with zero references
mcp_milens_grep	Text search for tech debt markers and dangerous patterns

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

Workflow

Step 1: PR Overview

Start with the full PR-level assessment.

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

Review the output for:

• Changed files — what was touched
• Affected symbols — each with a risk score (CRITICAL, HIGH, MEDIUM, LOW)
• Risk summary — distribution of risk levels across the change

Step 2: Deep-Dive on High-Risk Symbols

For every symbol rated CRITICAL or HIGH, perform a deep-dive.

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

What you get:

• Role — what the symbol does in the architecture
• Heat — frequency of changes, complexity, fan-in/fan-out
• Dependents count — how many things break if this is wrong
• Test status — whether it has test coverage

Step 3: Context for High-Risk Symbols

Still for CRITICAL/HIGH symbols, get 360° context.

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

This reveals:

• Incoming references — who depends on this symbol
• Outgoing calls — what this symbol depends on
• Re-export chains — if the symbol is re-exported through barrel files

Step 4: Dead Code Audit

Check for exported symbols with zero references (potential dead code).

mcp_milens_find_dead_code({repo: "<workspaceRoot>", limit: 20})

Flag any dead code that is:

• Newly introduced by this PR
• In files touched by this PR
• Adjacent to changed code (same module)

Step 5: Tech Debt Search

Search for tech debt markers and debugging remnants.

mcp_milens_grep({pattern: "TODO|FIXME|HACK|console\\.(log|debug)", scope: "code", repo: "<workspaceRoot>"})

Also check for:

• Commented-out code
• @ts-ignore or @ts-expect-error (TypeScript)
• # noqa or # type: ignore (Python)
• Empty catch blocks

Step 6: Produce Review Report

Consolidate findings into a review report with these sections:

1. Summary — files changed, risk distribution
2. Symbol Deep-Dives — one entry per CRITICAL/HIGH symbol with findings from Steps 2-3
3. Dead Code — symbols flagged from Step 4, with recommendations
4. Tech Debt — markers found in Step 5, with severity
5. Recommendations — prioritized list of actions (blocking vs. non-blocking)
6. Verdict — APPROVED / NEEDS CHANGES / BLOCKED

Example Session

Input

"review the PR for branch feature/auth-refactor"

Tool Calls

Step 1 — PR overview:

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

Output:

Changed files (4):
  src/auth/login.ts          [modified]
  src/auth/tokens.ts         [new]
  src/auth/types.ts          [modified]
  src/middleware/auth.ts     [modified]

Affected symbols:
  authenticate    [function]  CRITICAL  — 15 dependents, no tests
  generateToken   [function]  HIGH      — 8 dependents, 1 test
  UserSession     [interface] MEDIUM    — 5 dependents
  validateRole    [function]  LOW       — 2 dependents

Step 2 — Deep-dive on CRITICAL symbol:

mcp_milens_review_symbol({name: "authenticate", repo: "/home/user/project"})

Output:

Symbol: authenticate [function] — CRITICAL
  Role: core auth entry point, all protected routes depend on it
  Heat: 23 commits in last 90 days, 87% modification rate
  Dependents: 15 (12 direct callers + 3 middleware wrappers)
  Test status: 0 tests, 0% coverage
  Complexity: high (cyclomatic 14, 5 branching paths)

Step 3 — Context:

mcp_milens_context({name: "authenticate", repo: "/home/user/project"})

Output: 15 incoming refs (routes, middleware), 4 outgoing deps (token validation, user lookup, logger, rate limiter)

Step 4 — Dead code:

mcp_milens_find_dead_code({repo: "/home/user/project", limit: 20})

Output: legacyAuthCheck in the same module has 0 references — candidate for removal.

Step 5 — Tech debt:

mcp_milens_grep({pattern: "TODO|FIXME|HACK|console\\.(log|debug)", scope: "code", repo: "/home/user/project"})

Output: 3 matches — 1 console.log in tokens.ts, 2 TODO comments in login.ts.

Step 6 — Report produced (see report format above).

Best Practices

1. Don't skip the deep-dive. CRITICAL symbols demand review_symbol + context — surface-level review misses cascading failures.
2. Dead code in the diff is a red flag. If a new symbol has zero references, question whether it belongs in this PR.
3. Tech debt in new code is a blocker. console.log, FIXME, or debug code should not ship. Flag it as blocking.
4. Risk score drives review depth. HIGH/CRITICAL = full deep-dive. MEDIUM = quick context check. LOW = skip unless it's a new file.
5. Produce a written report. Don't just run the tools — consolidate findings into a review document the team can reference.

Quality Gate

Criteria	Pass	Fail
PR review tool	Review completes without errors	Tool fails or returns partial data
Risk assessment	All CRITICAL/HIGH symbols deep-dived	Any CRITICAL symbol skipped
Dead code check	No newly introduced dead code	New dead code found in changed files
Tech debt grep	No debug code (console.log, etc.)	Debug/leftover code in diff
Report completeness	All 6 report sections filled	Missing sections or incomplete findings