Simstim

# /simstim - HITL Accelerated Development Workflow

## Purpose

Orchestrate the complete Loa development cycle with integrated Flatline Protocol reviews at each stage. Human drives planning phases interactively while HIGH_CONSENSUS findings auto-integrate.

*"Experience the AI's work while maintaining your own consciousness."* — Gibson, Neuromancer

### Key Difference from /autonomous

| Aspect | /autonomous | /simstim |
|--------|-------------|----------|
| Designed for | AI operators (Clawdbot) | Human operators (YOU) |
| Planning phases | Minimal interaction, AI-driven | YOU drive interactively |
| Flatline results | BLOCKER halts workflow | BLOCKER shown to you, you decide |
| Implementation | Integrated into workflow | Hands off to /run sprint-plan |

## Getting Started

You can provide as much context as you want when invoking simstim:

```bash
# Simple invocation
/simstim

# With context — works great!
/simstim I want to build a user authentication system with OAuth2,
         JWT tokens, and role-based access control

# For large context, use the context directory
# Put files in grimoires/loa/context/ first, then:
/simstim

How It Works

Simstim guides you through 8 phases:

1. Phases 1-6 are interactive — you answer questions and make decisions
2. Phase 7 runs autonomously via /run sprint-plan
3. Each phase completes fully before the next begins

Note: Simstim has its own workflow structure and does NOT use Claude Code's Plan Mode.

Usage

# Full cycle from scratch
/simstim

# Skip to specific phase (requires existing artifacts)
/simstim --from architect       # Skip PRD (requires existing PRD)
/simstim --from sprint-plan     # Skip PRD + SDD
/simstim --from run             # Skip all planning, just run sprints

# Resume interrupted workflow
/simstim --resume

# Preview planned phases
/simstim --dry-run

# Abort and clean up
/simstim --abort

Flags

Flag	Description	Default
--from <phase>	Start from specific phase	-
--resume	Continue from interruption	false
--abort	Clean up state and exit	false
--dry-run	Show plan without executing	false

Flag Mutual Exclusivity

• --from and --resume cannot be used together
  • --from starts fresh from a phase (ignores existing state)
  • --resume continues from last checkpoint (requires existing state)
• --abort takes precedence over all other flags
• --dry-run can be combined with any flag

Phases

Phase	Name	Description
0	PREFLIGHT	Validate config, check state, beads health
1	DISCOVERY	Create PRD interactively
2	FLATLINE PRD	Multi-model review of PRD
3	ARCHITECTURE	Create SDD interactively
4	FLATLINE SDD	Multi-model review of SDD
5	PLANNING	Create sprint plan interactively
6	FLATLINE SPRINT	Multi-model review of sprint plan
6.5	FLATLINE BEADS	Iterative task graph refinement (v1.28.0)
7	IMPLEMENTATION	Autonomous execution via /run sprint-plan
8	COMPLETE	Summary and cleanup

Flatline Beads Loop (v1.28.0)

Phase 6.5 runs the "Check your beads N times, implement once" pattern when beads_rust is installed:

# Automatically triggered after FLATLINE SPRINT if:
# 1. beads_rust (br) is installed
# 2. Beads have been created from sprint tasks
# 3. flatline.beads_loop is enabled in config (default: true)

What Happens

1. Export: Current beads are exported to JSON
2. Review: Flatline Protocol reviews task graph for:
   • Granularity problems (tasks too large/vague)
   • Dependency issues (missing, cycles, ordering)
   • Completeness gaps (missing tasks)
   • Clarity problems (ambiguous acceptance criteria)
3. Apply: HIGH_CONSENSUS suggestions auto-integrate
4. Iterate: Repeat until changes < 5% for 2 consecutive iterations
5. Sync: Final state synced to git

Progress Display

FLATLINE BEADS LOOP
════════════════════════════════════════════════════════════

Iteration 1/6...
  HIGH_CONSENSUS: 3, DISPUTED: 1, BLOCKERS: 0
  Change: 15%

Iteration 2/6...
  HIGH_CONSENSUS: 1, DISPUTED: 0, BLOCKERS: 0
  Change: 8%

Iteration 3/6...
  HIGH_CONSENSUS: 0, DISPUTED: 0, BLOCKERS: 0
  Change: 2%

FLATLINE DETECTED
════════════════════════════════════════════════════════════
Task graph stabilized after 3 iterations.

Skip Conditions

The phase is skipped when:

• beads_rust not installed (silent skip)
• No beads created from sprint tasks
• simstim.flatline.beads_loop: false in config
• User chooses to skip when prompted

Flatline Integration (HITL Mode)

During Flatline review phases (2, 4, 6, 6.5), findings are categorized:

Category	Criteria	Action
HIGH_CONSENSUS	Both models >700	Auto-integrate (no prompt)
DISPUTED	Score delta >300	Present to you for decision
BLOCKER	Skeptic concern >700	Present to you for decision (NOT auto-halt)
LOW_VALUE	Both <400	Skip silently

DISPUTED Handling

DISPUTED: [suggestion]
GPT scored 650, Opus scored 350.

[A]ccept / [R]eject / [S]kip?

BLOCKER Handling

BLOCKER: [concern]
Severity: 750

[O]verride (requires rationale) / [R]eject / [D]efer?

If you choose Override, you must provide a rationale that is logged to the trajectory.

State Management

Simstim tracks progress in .run/simstim-state.json:

{
  "simstim_id": "simstim-20260203-abc123",
  "state": "RUNNING",
  "phase": "flatline_sdd",
  "phases": {
    "preflight": "completed",
    "discovery": "completed",
    "flatline_prd": "completed",
    "architecture": "completed",
    "flatline_sdd": "in_progress",
    ...
  },
  "artifacts": {
    "prd": {"path": "grimoires/loa/prd.md", "checksum": "sha256:..."},
    "sdd": {"path": "grimoires/loa/sdd.md", "checksum": "sha256:..."}
  }
}

Resuming After Interruption

If your session is interrupted ```