odd

Gate: SMALL vs SUBSTANTIAL

Criteria	SMALL (direct)	SUBSTANTIAL (task plan)
Lines	≤50L code; ≤150L docs-only 1-file	>50L or unknown
Files	1 file	>1 file
Schema/auth/API	None	Any
External deps	None	New deps
Tier	0 (known)	1+
Commits forecast	≤2	>2

All criteria must pass for SMALL. Otherwise → create odd/tasks/<feature>.md.

Slice Plan

1 slice = 1 conventional commit · ≤400L code (≤800L docs-only) per slice · tests in same commit as behavior.

Slice 1: <scope> (<est. lines>) — <outcome>
Slice 2: ...

Presets (ref: opencode-model-router)

Preset	Phase	Agent	Fallback
P1	Propose	analysis agent	vMK
P2	Slice plan	implementation agent	quick
P3	Implement	implementation agent	codex
P4	Verify	quality agent	deep
P5	Review	review agent	vMK

Workflow (ODD Cycle)

1. Gate — Evaluate SMALL vs SUBSTANTIAL criteria above.
2. Slice — Break into ≤400L units. Each slice gets odd/tasks/<feature>.md entry.
3. Implement — One preset per phase. Commit per slice conventional.
4. Verify — Per-slice tests pass + repo works after each commit.
5. Review — Log review per slice in task plan Review Log.

Rollback

Each slice commits independently. Revert slice N without touching N-1 or N+1.

Anti-Rationalization

Rationalization	Red Flag	Verification
"One big commit is fine"	>400L or >1 logical unit in one commit	Slice Plan: 1 commit = 1 deliverable outcome
"Skip review for small"	No review log entry	Workflow step 5 mandatory per slice
"Reuse SDD registry"	Confusing ODD slices with SDD spec tracking	ODD=slices+review, SDD=qué/cuándo registry

Red Flags

• Slice >400L → split further
• No test in commit with behavior → STOP, add test
• Preset skip without fallback → route through opencode-model-router

Verification

• cross-ref-check.ps1 → SKILL.md OK
• Each commit: git diff --stat ≤400L

Refs

Cross-Refs: work-unit-commits | sdd-quick | commit-crafter | opencode-model-router

Reference

Template + examples + presets detail → {file:odd/reference.md}