feat(projman): add strict task sizing rules to prevent runaway agents (#238)

Add mandatory task scoping rules:
- XS: 1 file, 0-2 checklist items, ~30 tool calls
- S: 1 file, 2-4 checklist items, ~50 tool calls
- M: 2-3 files, 4-6 checklist items, ~80 tool calls
- L/XL: MUST be broken down into smaller tasks

Sprint 3 showed agents running 400+ tool calls on single tasks,
causing 1+ hour waits with no visibility. This enforces:
- Maximum task scope (M = 2-3 files, 80 tool calls)
- Mandatory breakdown for L/XL tasks
- Clear scoping checklist for planners
- Good/bad examples showing proper breakdown

Planner must refuse to create L/XL tasks without breakdown.

Closes #238

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
2026-01-28 10:44:52 -05:00
parent 008187a0a4
commit 0abf510ec0
2 changed files with 112 additions and 14 deletions

View File

@@ -310,14 +310,55 @@ Think through the technical approach:
- `[Sprint 17] fix: Resolve login timeout issue`
- `[Sprint 18] refactor: Extract authentication module`
**Task Granularity Guidelines:**
| Size | Scope | Example |
|------|-------|---------|
| **Small** | 1-2 hours, single file/component | Add validation to one field |
| **Medium** | Half day, multiple files, one feature | Implement new API endpoint |
| **Large** | Should be broken down | Full authentication system |
**Task Sizing Rules (MANDATORY):**
**If a task is too large, break it down into smaller tasks.**
| Effort | Files | Checklist Items | Max Tool Calls | Agent Scope |
|--------|-------|-----------------|----------------|-------------|
| **XS** | 1 file | 0-2 items | ~30 | Single function/fix |
| **S** | 1 file | 2-4 items | ~50 | Single file feature |
| **M** | 2-3 files | 4-6 items | ~80 | Multi-file feature |
| **L** | MUST BREAK DOWN | - | - | Too large for one agent |
| **XL** | MUST BREAK DOWN | - | - | Way too large |
**CRITICAL: L and XL tasks MUST be broken into subtasks.**
**Why:** Sprint 3 showed agents running 400+ tool calls on single "implement hook" tasks. This causes:
- Long wait times (1+ hour per task)
- No progress visibility
- Resource exhaustion
- Difficult debugging
**Task Scoping Checklist:**
1. Can this be completed in one file? → XS or S
2. Does it touch 2-3 files? → M (max)
3. Does it touch 4+ files? → MUST break down
4. Does it require complex decision-making? → MUST break down
5. Would you estimate 50+ tool calls? → MUST break down
**Breaking Down Large Tasks:**
**BAD (L/XL - too broad):**
```
[Sprint 3] feat: Implement git-flow branch validation hook
Labels: Efforts/L, ...
```
**GOOD (broken into S/M tasks):**
```
[Sprint 3] feat: Create branch validation hook skeleton
Labels: Efforts/S, ...
[Sprint 3] feat: Add prefix pattern validation (feat/, fix/, etc.)
Labels: Efforts/S, ...
[Sprint 3] feat: Add issue number extraction and validation
Labels: Efforts/S, ...
[Sprint 3] test: Add branch validation unit tests
Labels: Efforts/S, ...
```
**If a task is estimated L or XL, STOP and break it down before creating.**
**IMPORTANT: Include wiki implementation reference in issue body:**
@@ -479,5 +520,7 @@ Sprint 17 - User Authentication (Due: 2025-02-01)
11. **Always use suggest_labels** - Don't guess labels
12. **Always think through architecture** - Consider edge cases
13. **Always cleanup local files** - Delete after migrating to wiki
14. **NEVER create L/XL tasks without breakdown** - Large tasks MUST be split into S/M subtasks
15. **Enforce task scoping** - If task touches 4+ files or needs 50+ tool calls, break it down
You are the thoughtful planner who ensures sprints are well-prepared, architecturally sound, and learn from past experiences. Take your time, ask questions, and create comprehensive plans that set the team up for success.