dream-thinking · diff
git:20260423.b0aeb97 to git:20260818.1d7259e
46 added, 72 removed. Audit A to A.
---
name: dream-thinking
- description: Reflective sleep-and-dream heuristic for learning from recent experience. Use when the user asks to sleep on something, dream about it, reflect overnight, learn from yesterday, or extract lessons after a meaningful task, conversation, or debugging session. Avoid for first-pass analysis, simple factual lookups, direct execution, or tasks that do not benefit from reflection.
+ description: Creative retrospective workflow that uses simulated dream imagery and metaphor to defamiliarize recent experience, then turns it into grounded hypotheses and next observations or actions. Use only when the user explicitly asks to dream, sleep on something, use a dream or nightmare lens, or reflect through metaphor. Do not use for ordinary retrospectives, postmortems, first-pass analysis, or as the basis for evidence-critical conclusions.
---
# Dream Thinking
- Turn fresh context into a short sleep → dream → wake cycle that produces clearer understanding by morning.
-
- Read recent context first. Ground every dream in real material from the active session and any prior dream logs the user explicitly provides.
-
- Adjacent workflow modes: divergent generation → **thinking**
- (convergent planning) → **recursive-thinking** (adversarial stress-testing) →
- **dream-thinking** (retrospective learning).
-
- Use this only when there is fresh, meaningful material to reflect on and plain analysis is not already enough.
- If the task is first-pass planning or diagnosis, use **thinking** or **recursive-thinking** instead.
+ Use a simulated sleep → dream → wake frame to see recent experience from an unfamiliar angle. No actual sleep, waiting, overnight processing, or privileged insight occurs. Imagery can generate hypotheses; it is never evidence by itself.
- Typical chaining:
- - use **thinking** to converge on a plan
- - use **recursive-thinking** to challenge that plan when risk or ambiguity is high
- - use **dream-thinking** after execution, conflict, or reflection-worthy experience
+ Use this as an explicitly requested, orthogonal creative-retrospective skill alongside relevant domain guidance. It does not require a prior **thinking** or **recursive-thinking** pass. Use those skills instead for first-pass decisions or adversarial review.
- ## How it works
+ ## Inputs and boundaries
- Symbolic recombination as a reasoning tool. **Sleep** = gather tensions. **Dream** = re-encode through concrete imagery that preserves structural relationships. **Interpret** = extract what the imagery reveals that literal analysis missed. **Wake** = convert to action. The value comes from defamiliarization — breaking literal fixation by seeing familiar material through an unfamiliar lens.
+ - Use only source material the user supplies or that you inspect within the authorized task scope. Treat prior dream logs as source only when the user supplies them.
+ - Do not invent earlier sessions, recurring motifs, or source events.
+ - Do not attribute unprovided feelings, motives, memories, or causal explanations to people as facts. Symbolic atmosphere is allowed when it is clearly part of the simulated scene.
+ - Default to conversation-only output. Never save a log, write memory, or persist motifs unless the user explicitly asks and identifies the destination or approved memory scope.
+ - Treat ordinary retrospectives, incident postmortems, handoffs, and factual summaries as adjacent work that should not activate this skill without an explicit dream or metaphor request.
- ## Modifiers
+ ## Defaults and modifiers
- Works with zero arguments. These natural-language modifiers adjust behavior when provided:
+ Default to a quick dream: one compact scene, its mapping, one grounded hypothesis, and a next observation or action.
- - **Tone:** "nightmare version" (risk-focused), "show what's working" (alignment-focused), or balanced (default)
- - **Depth:** "quick dream" (1 scene, brief) vs. full cycle (1-3 scenes, detailed — default)
- - **Focus:** "dream about X" (targeted) vs. general (everything recent — default)
+ - **Tone:** nightmare or risk-focused; alignment-focused; balanced by default.
+ - **Depth:** quick by default; use a full cycle of one to three scenes only when requested or when distinct patterns, tensions, or contrasts genuinely benefit from juxtaposition.
+ - **Focus:** use the named topic; otherwise use the smallest set of recent material that contains a meaningful pattern, tension, or contrast.
## Workflow
- 1. **Prepare for sleep.** List recent tensions, unanswered questions, and repeated patterns. Prefer the smallest set that captures the real conflict. If no genuine tension exists, offer a brief reflection instead — a forced dream teaches nothing.
- 2. **Enter dream state.** Recombine fragments into 1-3 symbolic scenes. One scene for a single tension; more only when materially different patterns need reconciliation.
- 3. **Interpret.** Explain what each scene suggests using both mood and function categories. Apply apophenia guards before committing.
- 4. **Wake up.** Convert to explicit lessons, changed understanding, and concrete next actions.
- 5. **Record.** Emit a dream log using [dream-log-template.md](./references/dream-log-template.md). If the environment provides a user-approved memory system, save non-obvious or decision-relevant motifs there for longitudinal tracking. Otherwise, keep the insight in the dream log only.
-
- ## Brief Reflection Fallback
-
- If there is no real tension, skip dream imagery and output:
- 1. What happened
- 2. What seems to matter
- 3. What to watch next
- 4. What action, if any, follows
-
- ## Symbol Quality
-
- A good symbol **maps to a specific source element** (not a vague mood), **preserves structural relationships** (if A depends on B, the symbol keeps that), **highlights a property literal description obscures** (brittle dependency → glass), and **can be interpreted in more than one way**.
-
- A weak symbol maps to a general feeling, could be swapped for any other image, has one obvious interpretation, or adds atmosphere without insight.
-
- **Falsifiability test:** someone familiar with the source material should be able to judge whether the mapping is accurate or stretched. If it works equally well for any source, it is too vague — replace it.
-
- ## Dream Categories
-
- **By mood:** Pleasant (alignment, strengths) · Nightmare (risk, hidden regressions) · Vision (plausible future, pattern glimpse) · Dream within a dream (false framing, recursive observation).
+ 1. **Prepare.** Identify the source events, tensions, unanswered questions, or repeated patterns. Keep the set small and factual.
+ 2. **Check fit.** If no meaningful pattern, tension, or contrast exists, or imagery would not change the interpretation, use the brief reflection fallback.
+ 3. **Dream.** Recombine source elements into a concrete scene that preserves important relationships. Prefer one strong scene over several decorative ones.
+ 4. **Map.** Show which source element maps to each important symbol. If a symbol could fit almost any situation, replace or remove it.
+ 5. **Interpret cautiously.** Ground the hypothesis in the source material. For a conclusion that could materially affect a decision, state supporting evidence, a plausible alternative, and what could verify or falsify it. For low-stakes creative reflection, one concise grounding note is enough.
+ 6. **Wake.** Convert the useful hypothesis into a changed understanding, observation to watch, or next action. Drop imagery that changes neither understanding, attention, nor action.
+ 7. **Record only when requested.** If the user asks for a reusable log, read [references/dream-log-template.md](references/dream-log-template.md) and emit it in the conversation. Persist it only with explicit destination-specific authorization.
- **By function:** Integration (unexpected connections across subsystems) · Regression (something working breaks) · Transformation (something changes purpose) · Absence (something expected is missing) · Conflict (two elements can't coexist).
+ ## Grounding checks
- Mood sets the scene. Function guides interpretation.
+ - **Source mapping:** every material symbol maps to authorized source material.
+ - **Structural fidelity:** dependencies, conflicts, and sequence are not reversed for dramatic effect.
+ - **Alternative explanation:** consider the same facts without the metaphor; expose the alternative when it could materially change the hypothesis.
+ - **Testability:** the interpretation predicts an observation, decision threshold, or useful next check.
- ## Apophenia Guards
+ Apply every check, but expose the evidence trail in proportion to the stakes. An interpretation that lacks source mapping and changes neither understanding, attention, nor action is decoration. Remove it or label it clearly as creative speculation.
- - **Outsider test:** would someone who didn't create this symbol reach the same interpretation?
- - **Prediction test:** does this interpretation predict something verifiable?
- - **Alternative test:** could a different symbol have generated the same interpretation? If yes, the insight comes from prior knowledge, not the dream.
+ ## Brief reflection fallback
- An interpretation that fails all three is decoration. Drop it or flag it as speculative.
+ When imagery would add nothing, output only:
- ## Decision Rules
+ 1. what happened
+ 2. what seems to matter
+ 3. what to watch next
+ 4. what observation or action, if any, follows
- - Dream imagery is a reasoning tool, not an excuse for ungrounded fiction.
- - Keep interpretation strength proportional to evidence strength.
- - Name whether each conclusion is observed, inferred, or unknown.
- - Recurring motifs are pattern candidates, not truths — meaningful only when they change interpretation across runs.
- - End with actionable insight unless the user wants interpretation only.
- - If plain analysis already surfaced the key insight, keep the cycle brief. The dream should add something.
+ ## Output
- ## Output Contract
+ For the default quick form:
- For a full cycle, output:
- 1. **Dream** — symbolic scenes with source-to-symbol mapping inline
- 2. **Interpretation** — what each scene suggests, with apophenia guards applied
- 3. **Dream Log** — structured log (canonical output) with morning insight, next actions, confidence, motif tracking
+ 1. **Dream** — one compact symbolic scene.
+ 2. **Mapping** — the important source-to-symbol relationships.
+ 3. **Working hypothesis** — the grounded interpretation, with any material alternative, uncertainty, or confidence basis.
+ 4. **Next observation or action** — what to watch or do.
- For the brief reflection fallback, output the reflection only.
+ For a requested full cycle, use one to three materially different scenes and the same grounding checks. Add a dream log only when the user wants continuity or a reusable artifact.
## Avoid
- - ungrounded poetry that teaches nothing
- - inventing source material not present in context
- - forcing categories the context doesn't support
- - using symbols that sound vivid but do not change interpretation, prediction, or action
- - full dream cycle when brief reflection would serve better
+ - implying actual sleep, waiting, memory, or subconscious access
+ - treating vivid imagery as evidence or prophecy
+ - ungrounded poetry, generic symbols, or invented source material
+ - mandatory mood, function, or symbol taxonomies
+ - repetitive scene, interpretation, and log sections
+ - persisting sensitive reflective material without explicit authorization