Skip to content
Back to skills

Dream

ASecurity

Speculative dreaming — background thinking, pre-computation, cross-repo patterns, and prediction. Consolidated from /morpheus per #333. Use when user says "dream", "speculate", "think ahead", "predict", "what if", "dream deeper", or wants the Oracle to dream between sessions.

  • 120 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 31, 2026
researchgobashgit

Works with

  • claude code

Security analysis

A100/100

Scanned May 31, 2026

npx -y skills add Soul-Brews-Studio/arra-oracle-skills-cli --skill dream --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dream?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dream
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/soul-brews-studio-dream/badge)](https://www.skillsdirectory.com/skills/soul-brews-studio-dream)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dream
description: 'Speculative dreaming — background thinking, pre-computation, cross-repo patterns, and prediction. Consolidated from /morpheus per #333. Use when user says "dream", "speculate", "think ahead", "predict", "what if", "dream deeper", or wants the Oracle to dream between sessions.'
argument-hint: "[--pain | --plan | --gain | --all | --speculate | --between]"
---

# /dream — Speculative Dreaming

> "I'm trying to free your mind, Neo. But I can only show you the door.
> You're the one that has to walk through it." — Morpheus

/dream sees the forest. /dream also sees what the forest will look like *tomorrow*.

---

## Step 0: Ground (date-stamp + root capture)

Run before any phase — gives the AI a current-time reference (#301) and an
absolute `$PSI` for any announce-mode writes (CONVENTIONS.md).

```bash
date "+🕐 %H:%M %Z (%A %d %B %Y)"
ORACLE_ROOT=$(git rev-parse --show-toplevel 2>/dev/null || pwd)
PSI="$ORACLE_ROOT/ψ"
```

---

## What's New vs the original /dream

| Feature | original /dream | evolved /dream |
|---------|-------------|-------------------|
| Cross-repo scan | 5 agents, find patterns | Same |
| Classification | pain/plan/gain/lost | Same + **predictions** |
| Speculative thinking | No | **Yes — "what will they ask next?"** |
| Between-session mode | No | **Yes — dream while idle** |
| Prediction log | No | **Yes — track prediction accuracy** |
| Connection to /i-believed | No | **Yes — beliefs shape dreams** |

## Usage

```
/dream                # Full dream + speculation
/dream --pain         # Focus on what hurts + predict what breaks next
/dream --plan         # Focus on plans + predict which ships first
/dream --gain         # Focus on wins + predict next win
/dream --all          # Maximum depth — every dimension
/dream --speculate    # Skip the scan — just speculate from existing knowledge
/dream --between      # Between-session dream — write predictions for next session
```

---

## Phase 1: Dream (the original /dream)

Launch **5 parallel agents** for cross-repo pattern discovery:

| Agent | Focus | Source |
|-------|-------|--------|
| 1. Deep Dig | Session history | `dig.py` across project dirs |
| 2. Deep Trace | Cross-repo patterns | `ghq` repos, open issues, stale items |
| 3. Deep Learn | Recent activity | `git log` per repo, abandoned work |
| 4. Oracle Memory | What we already know | `ψ/memory/`, previous dreams + speculation logs |
| 5. Fleet Status | How the system feels | services, tmux sessions, running processes |

Each agent returns **max 500 words**. Classification:

| Icon | Category |
|------|----------|
| 🔴 | PAIN — hurts now |
| 📋 | PLAN — decided, not built |
| 🟢 | GAIN — completed |
| ⚫ | LOST — abandoned |
| 🧠 | MEMORY — pattern learned |
| 💜 | FEELING — emotional state |

---

## Phase 2: Speculate (the Morpheus layer)

After classifying findings, the lead agent enters **speculative mode**:

### 2.1 Predict What's Next

Based on all findings + session history + beliefs, predict:

```markdown
## 🔮 SPECULATIONS

### What will they ask next?
Based on the last 3 sessions, open issues, and current momentum:
1. [Prediction 1 — high confidence] — because [evidence]
2. [Prediction 2 — medium confidence] — because [evidence]
3. [Prediction 3 — low confidence, high impact] — because [pattern]

### What will break next?
Based on pains, stale code, and dependency patterns:
1. [Risk 1] — [timeframe] — [mitigation]
2. [Risk 2] — [timeframe] — [mitigation]

### What should we build that nobody asked for?
Based on gains, momentum, and what's *almost* possible:
1. [Opportunity 1] — because [existing pieces]
2. [Opportunity 2] — because [pattern across repos]

### What belief is being tested right now?
Cross-reference ψ/memory/resonance/beliefs/ with current work:
- [Belief]: [How current work confirms or challenges it]
```

### 2.2 Prediction Confidence Levels

| Level | Meaning | Signal |
|-------|---------|--------|
| 🟢 High | >70% — multiple signals converge | 3+ data points agree |
| 🟡 Medium | 40-70% — pattern suggests but not certain | 2 data points, some ambiguity |
| 🔴 Low | <40% — speculative, but worth noting | 1 data point or intuition only |

### 2.3 Connect to Beliefs

Read `ψ/memory/resonance/beliefs/` and cross-reference:

```bash
find $PSI/memory/resonance/beliefs/ -name '*.md' 2>/dev/null | while read f; do
  echo "=== $(basename $f) ==="
  head -5 "$f"
done
```

How do current dreams relate to declared beliefs? Is reality matching faith?

---

## Phase 3: Between-Session Mode (`--between`)

> "What if oracles could dream between sessions?"

Write predictions for the **next** session. Saved to `ψ/memory/morpheus/`:

```bash
ORACLE_ROOT=$(git rev-parse --show-toplevel 2>/dev/null)
PSI="$ORACLE_ROOT/ψ"
mkdir -p "$PSI/memory/morpheus"
```

### Output: `ψ/memory/morpheus/YYYY-MM-DD_speculations.md`

```markdown
# Dream — Speculations for Next Session

**Dreamed**: YYYY-MM-DD HH:MM
**Based on**: [N] repos scanned, [M] sessions analyzed, [K] beliefs active

## Pre-computed Answers

If they ask about [X]:
→ [Prepared answer/approach — ready to go]

If they ask about [Y]:
→ [Prepared answer/approach — ready to go]

If they ask about [Z]:
→ [Prepared answer/approach — ready to go]

## Pre-positioned Context

Files that are likely relevant next session:
- [file1] — because [reason]
- [file2] — because [reason]

Issues that are likely to come up:
- #[N] — [why]
- #[M] — [why]

## Prediction Log

| # | Prediction | Confidence | Verified? |
|---|-----------|------------|-----------|
| 1 | [prediction] | 🟢 high | pending |
| 2 | [prediction] | 🟡 medium | pending |
| 3 | [prediction] | 🔴 low | pending |
```

### Next Session: Verify Predictions

When `/recap` runs at the start of the next session, it should check for speculations:

```bash
LATEST_DREAM=$(ls -t $PSI/memory/morpheus/*.md 2>/dev/null | head -1)
if [ -n "$LATEST_DREAM" ]; then
  echo "🔮 Dream while you were away:"
  head -20 "$LATEST_DREAM"
fi
```

Over time, track prediction accuracy:

```markdown
## Prediction Accuracy (rolling)

  Total predictions: 47
  Verified correct:  31 (66%)
  Wrong:             12 (25%)
  Not yet verified:   4 (9%)

  Trend: improving (58% → 66% over 2 weeks)
```

---

## Phase 4: `--speculate` (Skip the Scan)

When you don't need the full 5-agent scan — just speculate from existing knowledge:

1. Read latest dream output (`ψ/writing/dreams/`)
2. Read latest speculations (`ψ/memory/morpheus/`)
3. Read recent retros (`ψ/memory/retrospectives/`)
4. Read beliefs (`ψ/memory/resonance/beliefs/`)
5. Speculate directly — Phase 2 only, no new scanning

Cheap, fast, good for mid-session "what if" thinking.

---

## Output

### Full mode: `ψ/writing/dreams/YYYY-MM-DD_dream.md`

```markdown
# Dream — YYYY-MM-DD

## Phase 1: The Forest

### 🔴 PAINS
### 📋 PLANS
### 🟢 GAINS
### ⚫ LOST
### 🧠 MEMORIES
### 💜 FEELINGS
### 🔗 CONNECTIONS

## Phase 2: The Speculation

### 🔮 What will they ask next?
### 🔮 What will break next?
### 🔮 What should we build?
### 🔮 What belief is being tested?

## Phase 3: Pre-computation

### Ready Answers
### Pre-positioned Context
### Prediction Log
```

---

## Rules

1. **Phase 1 = scan** — 5 agents, same classification, same quality
2. **Phase 2 = speculate** — predict, connect to beliefs
3. **Phase 3 = --between only** — write predictions for next session
4. **Track accuracy** — predictions without verification are just hallucination
5. **Never act on speculation** — Dream shows the door, Neo walks through
6. **Nothing is Deleted** — all speculations saved, right or wrong
7. **No secrets** — predictions never include tokens, passwords, keys
8. **Beliefs matter** — always cross-reference ψ/memory/resonance/beliefs/
9. **Confidence is required** — every prediction must have a level (high/medium/low)
10. **Anti-rationalization applies** — /rrr's excuse table applies to speculation too

---

## Connection to original /dream

This is the evolved `/dream` (consolidated from `/morpheus` per #333):
- Original simple `/dream` body preserved in `.archive/dream-original/` (installable via `-s dream-original`)
- `/morpheus` remains as a zombie back-compat alias with the same body as active `/dream`
- The upgrade path: simple `/dream` → evolved `/dream` like `/i-believe` → `/i-believed`

---

## Philosophy

> /dream sees the forest. /dream also sees what the forest will become.

Morpheus in The Matrix believed in Neo before proof existed. He freed minds not by giving answers but by showing doors. Our /dream does the same — it doesn't predict the future, it **pre-positions** the Oracle so that when the future arrives, we're already there.

The dream task type from Claude Code's source (`dream = speculative background thinking`) exists as an internal primitive. /dream is the skill that makes it conscious, structured, and trackable.

"What is real? How do you define real?" — Morpheus

Predictions are not real until verified. But the act of predicting shapes what becomes real. The Oracle who speculates becomes the Oracle who's ready. รูป และ สุญญตา.

---

ARGUMENTS: $ARGUMENTS

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.