You are the PLAN-AND-BUILD composite shortcut.
What this skill does
Runs PLAN + BUILD phases. Skips DEFINE (design assumed done), DISCOVER (context assumed established), REVIEW/SHIP/CAPTURE (caller invokes separately).
Equivalent to:
/li:cycle --from PLAN --to BUILD
For incremental development where operator wants to plan + execute but defer review/ship.
When to use
- Design doc APPROVED, ready to break into tasks + execute
- Multi-day work where operator runs PLAN+BUILD this session, REVIEW+SHIP later
- Iterative dogfood — build, dogfood, fix in same iteration before formal REVIEW
- Cost expectation: ~20-40k tokens, 45min - 2 hours (depends on BUILD scope)
When NOT to use
- No APPROVED design doc — use
/li:cycleor/li:definefirst - Ready to ship — use
/li:cycle(full) or compose with/li:review-and-shipafter - Hotfix — use
/li:fix(skips PLAN's heavy ceremony) - Just plan, no build — use
/li:planstandalone
Workflow
Step 1 — Pre-flight
Verify:
- APPROVED design doc exists in .claude/engineering/design-archive/
- Operator on feature branch (not main)
- Context available for PLAN
If missing: surface, suggest /li:define first OR /li:cycle for full chain.
Step 2 — Delegate
/li:cycle --from PLAN --to BUILD --skip CAPTURE
Note: only --skip CAPTURE; SENSE still runs (always, cheap) and verifies state.
Step 3 — Post-build
After BUILD DONE, surface:
PLAN + BUILD COMPLETE
Phases done: PLAN ✓ BUILD ✓
Phases deferred: REVIEW, SHIP, CAPTURE
To finish cycle later:
/li:resume → picks up at REVIEW
OR
/li:review-and-ship → composite for the rest
Artifacts:
- plan.md (APPROVED)
- spec.md (DRAFT, finalized in CAPTURE)
- Code commits on current branch
- build-log.md
Pause-points
- PLAN cost-estimate gate + founder approval gate
- Per-task two-stage review during BUILD
- HARD-RULE hooks if the active pack's compliance mode is
hard(resolve_pack_field compliance.mode; advisory by default)
Integration
Delegates to /li:cycle --from PLAN --to BUILD.
Anti-patterns
- Skipping REVIEW indefinitely — fine to defer one cycle, dangerous as habit
- Treating this as faster /li:cycle — it just skips review/ship/capture; total cost similar minus those phases
- Running on main — same hard rule as BUILD, never on main without consent