execution-mode

When to Use

Auto-detect task execution mode — QUICK/THOROUGH/DRAFT — from scope, risk, familiarity. Not resource optimization.

Modes

Mode	When	Depth	Verification
QUICK	Simple bugfix, known	Minimal	Tests only
THOROUGH	Complex, risky	Full SDD	Full gate
DRAFT	Explore, prototype	Notes	Skip
User says "careful"/"validate" on QUICK → escalate THOROUGH. User override wins.

Context Zones (via context-watchdog)

GREEN <40%: full depth | YELLOW 40-60%: L1+L2 | ORANGE 60-80%: L2+L3 compact@70% | RED >80%: mem_save+break. Re-evaluate every 5 tools. Escalate on any HIGH. De-escalate after 3 lower. User override wins.

Mode Pipeline

Mode	Pipeline	Output
QUICK	sdd-quick (Propose→Apply→Verify)	Code+tests. Score+move.
THOROUGH	Full SDD 9-phase	Every decision→Engram. Quality gate. PR with evidence.
DRAFT	Explore first	Notes. No commit without user OK.
THOROUGH: Init→Propose→Spec→Design→Tasks→Apply→Verify→Quality Gate→Archive. Every step→mem_save checkpoint.

Anti-Patterns

DRAFT when asked THOROUGH · Skip zone re-eval · Ignore RED · One mode all session · Auto-detect without explaining · THOROUGH for single-line typo · DRAFT for destructive ops

docs/skills/execution-mode/reference.md

Anti-Rationalization

Rationalization	Red Flag	Verification
"QUICK para todo porque es más rápido"	Bypassing THOROUGH on risky/complex task	Risk/scope gate — complex/risky→THOROUGH 9-phase + quality-gate
"Un modo toda la sesión"	No zone re-eval after 5 tools	Re-evaluate every 5 tools — GREEN/YELLOW/ORANGE/RED dictates depth
"DRAFT no necesita permiso"	Committing from DRAFT without user OK	DRAFT→notes only — verify no commit without explicit user approval

Red Flags

• Doing work without checking output format → STOP, re-read skill
• Second occurrence of same rationalization → force RED zone

Verification

• Output matches skill ## Output contract + file:line citaton
• cross-ref-check.ps1 → SKILL.md OK

Refs

Cross-Refs: sdd | sdd-apply | development-mode