checklist

Pre-Implementation Checklist

This is not a checkbox exercise. This is your implementation blueprint — a real analysis document that you fill with actual research before writing any code.

When to Use

Before editing ANY source code file (.py, .ts, .tsx, .js, .jsx, .go, .rs, .java, .c, .cpp), you must create .sentinel/current-task.json. The pre-edit-gate hook will block edits without it.

The Checklist File

Create .sentinel/current-task.json with ALL of these fields:

{
  "task_id": "UNIQUE-ID",
  "why": "Business/quality reason for this change",
  "approach": "Chosen approach with reasoning (Option A because X, rejected Option B because Y)",
  "impact_files": [
    "path/to/file.py:function_name — description of impact",
    "path/to/other.ts:Component — imports changed module"
  ],
  "blast_radius": {
    "tests_break": ["test_name — why it breaks"],
    "tests_add": ["new_test_name — what it verifies"]
  },
  "verify_command": "pytest tests/test_module.py -x && npm run build"
}

How to Fill Each Field (with real analysis, not placeholders)

task_id

A unique identifier for this task. Use your project's convention (CRIT-1, FEAT-42, BUG-7) or create a descriptive one (fix-login-crash, add-rate-limit).

why

Not "because the audit said so." State the actual business or quality reason:

• BAD: "Fix bug" / "Audit item" / "Needs fixing"
• GOOD: "Users get 500 error when clicking Save because PMRequest.get() doesn't exist on dataclasses"
• GOOD: "API endpoint leaks internal paths in error messages, violating OWASP information disclosure rules"

approach

Pick ONE approach, explain why, and document what you rejected:

1. Research the options:

   • Read the current code
   • Check documentation for the framework/library
   • Consider 2-3 different ways to fix it

2. Choose one and justify:

   "Option A: Use attribute access (req.phase) instead of dict access (req.get('phase')).
    Reason: PMRequest is a dataclass, not a dict. Direct attribute access is correct.
    Rejected Option B: Convert PMRequest to dict first — unnecessary overhead, changes API contract.
    Band-aid check: This is root-cause fix — the dataclass was always a dataclass, the code was always wrong."