atlas-decisions

name	atlas-decisions
version	2.0
type	autoresearch
description	Strategic decision engine with multi-model consensus
user-invocable	true
agent	ATLAS
agent_model	claude-opus-4-6
mcps	["aegis","paperclip","command-center"]
gstack_skills	["/office-hours","/plan-ceo-review","/autoplan"]
eval_metric	decision_confidence_score
eval_budget	180s
guard	strategic_alignment > 0.7

/atlas-decisions — Strategic Decision Engine

What it does

Runs strategic decisions through multi-model consensus using AEGIS, real-world data via Paperclip, and gstack's CEO/strategy review pipeline.

Protocol

Step 0: Load context

• memory/CEO_MEMORY.md — strategic state, priorities, active decisions
• HENRY_BRAIN.md — dual-track strategy (Agency + Dark Factory)
• Current sprint from ENGINE

Step 1: Frame the decision

Every strategic decision gets structured as:

DECISION: [What needs to be decided]
CONTEXT: [Why now, what changed]
OPTIONS: [A, B, C — no more than 3]
CRITERIA: [What matters — revenue speed, risk, resource fit]
DEADLINE: [When this must be decided by]

Step 2: Gather intelligence

1. ORACLE provides research inputs (via oracle-dd if deal-related)
2. Paperclip fetches market data, competitor moves, industry news
3. LEDGER provides financial projections for each option
4. SHIELD flags legal/compliance risks per option

Step 3: Multi-model analysis (AEGIS)

Route the framed decision to AEGIS for diverse perspectives:

• Opus 4.6: Strategic reasoning, long-term implications
• DeepSeek R1: Quantitative analysis, probability assessment
• Gemini Flash: Speed analysis, quick-win identification

Synthesize: where do models agree? Where do they diverge? What does divergence signal?

Step 4: CEO review pipeline (/autoplan)

Run /autoplan which chains:

1. /plan-ceo-review — find the 10-star version
2. /plan-design-review — rate each dimension
3. /plan-eng-review — lock execution requirements

Step 5: Decision scorecard

DECISION SCORECARD
================================
Question:    [decision statement]
Deadline:    [date]

OPTION A: [name]
  Revenue speed:    /10
  Risk level:       /10 (10 = lowest risk)
  Resource fit:     /10
  Strategic align:  /10
  Model consensus:  [3/3, 2/3, 1/3]
  Total:            /40

OPTION B: [name]
  [same structure]

OPTION C: [name]
  [same structure]

RECOMMENDATION: [option] because [one sentence]
CONFIDENCE: [high/medium/low]
WHAT COULD CHANGE THIS: [key assumption]

Step 6: Route decision

• Log to command-center
• Update memory/CEO_MEMORY.md
• Dispatch execution to relevant agent:
  • Deal decision → CLOSER or SHIELD
  • Product decision → FORGE
  • Marketing decision → PULSE
  • Ops decision → ENGINE

Autoresearch Loop

• Metric: decision_confidence_score — composite of model consensus, data completeness, and option differentiation
• Budget: 180s eval
• Modify: decision framing templates, scoring weights, multi-model routing
• Guard: strategic_alignment with dual-track (Agency + Dark Factory) stays above 0.7

Active Strategic Decisions

Decision	Deadline	Status
TXS5345: Submit LOI or wait?	This week	Pending
Agency vs. acquisition: which first dollar?	This week	Pending
RIA registration: start now or after first acquisition?	This month	Pending
Fiverr/Upwork: worth the effort for bridge revenue?	This week	Pending

MCP Integration Map

ATLAS (Opus 4.6)
  ├── aegis: multi-model consensus (Opus + DeepSeek + Gemini)
  ├── paperclip: market data, competitor intel, industry news
  ├── command-center: log decisions, track outcomes
  ├── /office-hours: reframe strategic questions
  ├── /plan-ceo-review: find the 10-star answer
  └── /autoplan: full CEO → design → eng review pipeline