Initial commit

2025-11-30 08:59:09 +08:00
commit acc8e3e62e
9 changed files with 870 additions and 0 deletions
--- a/commands/strategize.md
+++ b/commands/strategize.md
@@ -0,0 +1,207 @@
+---
+name: strategize
+description: Analyze tasks through epistemic primitives, map to reasoning modalities, and execute skills with productive interleaving and full primitive analysis
+---
+
+# /strategize
+
+Design and execute a cognitive strategy with visible primitive analysis and modality interleaving. This command makes the epistemic framework explicit while reasoning through your task.
+
+## Execution Protocol
+
+When this command is invoked, follow this protocol precisely.
+
+### Phase 1: Elicit
+
+Use the `AskUserQuestion` tool to gather structured input about the task. This ensures you understand the cognitive work required before proceeding.
+
+**Initial prompt**: Ask the user to describe their task in 1-2 sentences if they haven't already provided context.
+
+**Structured elicitation**: Use `AskUserQuestion` to clarify:
+
+```
+AskUserQuestion with questions:
+1. "What type of cognitive work does this task primarily involve?"
+   Options:
+   - "Understanding/Investigation" (need to map territory, trace causes)
+   - "Creation/Design" (need to generate options, build something new)
+   - "Decision/Selection" (need to evaluate and choose between alternatives)
+   - "Problem-solving/Debugging" (need to find and fix what's wrong)
+   - "Synthesis/Integration" (need to combine disparate elements into coherent whole)
+
+2. "What constraints or context should guide the reasoning?"
+   Options:
+   - "Time-sensitive" (need efficient path to answer)
+   - "High-stakes" (need thorough validation)
+   - "Exploratory" (open to unexpected directions)
+   - "Bounded" (specific scope, don't expand)
+```
+
+After receiving answers, propose a session name based on the task (e.g., "api-redesign-strategy", "debugging-memory-leak"). This name unifies all artifacts created during the session.
+
+Proceed to Phase 2 once you have sufficient understanding of the task.
+
+### Phase 2: Analyze (Primitive Detection)
+
+Systematically evaluate which of the seven epistemic primitives the task demands. Present your analysis to the user with rationale for each:
+
+**FOCUS** (Allocate attention to specific elements)
+- Signals: "identify," "prioritize," "select," "narrow down," "most important," "isolate"
+- Strong need when: Task requires choosing what matters, filtering noise, concentrating on specifics
+
+**MOVE** (Navigate through epistemic space)
+- Signals: "explore," "navigate," "trace," "hierarchy," "layers," "causal chain," "follow"
+- Strong need when: Task involves traversing structures, following chains, moving between levels
+
+**LINK** (Create or discover relations)
+- Signals: "connect," "relate," "pattern," "dependencies," "how X affects Y," "relationship"
+- Strong need when: Task requires finding patterns, mapping dependencies, building connections
+
+**GENERATE** (Produce new possibilities)
+- Signals: "brainstorm," "alternatives," "options," "possibilities," "what if," "ideas," "create"
+- Strong need when: Task requires producing options, creative exploration, divergent thinking
+
+**EVALUATE** (Compare against criteria)
+- Signals: "assess," "validate," "check," "compare," "criteria," "which is better," "trade-offs"
+- Strong need when: Task requires judgment, validation, comparison, filtering options
+
+**TRANSFORM** (Convert between representations)
+- Signals: "restructure," "reframe," "simplify," "abstract," "concrete," "express differently"
+- Strong need when: Task requires changing form, shifting abstraction level, synthesizing
+
+**HOLD** (Maintain in working memory)
+- Signals: "maintain context," "complex," "juggling," "across steps," "multiple considerations"
+- Strong need when: Task has many moving parts, requires sustained attention across operations
+
+Display the primitive profile:
+```
+PRIMITIVE ANALYSIS:
+Your task requires these cognitive operations:
+  FOCUS     ████████░░  Narrowing to key components
+  MOVE      ██████░░░░  Navigating system structure
+  GENERATE  ████░░░░░░  Creating alternatives
+  EVALUATE  ████████░░  Comparing options
+  LINK      ██████░░░░  Mapping relationships
+  TRANSFORM ████░░░░░░  Restructuring understanding
+  HOLD      ██░░░░░░░░  Managing complexity
+```
+
+### Phase 3: Map (Modality Selection)
+
+Translate the detected primitives to modality requirements. Each modality is defined by its dominant primitive pair:
+
+**Exploratory** (FOCUS + MOVE)
+- Purpose: Navigate and map territory, follow threads, trace structures
+- Skills: tree-explore, graph-wander, root-cause, perspective-shift, domain-transfer, follow-intuition, trajectory-sense
+
+**Generative** (GENERATE + LINK)
+- Purpose: Create possibilities, produce options, divergent expansion
+- Skills: branch, wonder, scamper, alternatives, lateral-think, constraint-play, metaphor-build, scenario-planning, reversal
+
+**Evaluative** (EVALUATE + FOCUS)
+- Purpose: Filter, validate, assess, compare against criteria
+- Skills: evidence-check, logic-trace, weigh-options, fallacy-detect, opposite-case, confidence-calibrate, priority-matrix, leverage-find, fork-detect, premortem, decompose
+
+**Transformative** (TRANSFORM + LINK)
+- Purpose: Restructure, synthesize, change representation
+- Skills: abstraction-ladder, integrate, principle-extract, connect-dots, theme-find, feynman, first-principles, progressive-reveal
+
+**Introspective** (HOLD + FOCUS)
+- Purpose: Manage cognitive state, maintain context, monitor process
+- Skills: working-memory, attention-anchor, cycle-detect, knowledge-check, cognitive-load, thinking-monitor, keep-notes, reference-notes, session-reflect, self-observation, reconsider
+
+**Hybrid** (span multiple modalities)
+- Skills: dialectic, six-hats, socratic, systems-thinking
+
+Explain which modalities apply to this task and why, based on the primitive analysis.
+
+### Phase 4: Sequence (Skill Selection with Interleaving)
+
+Select skills based on the primitive requirements and arrange them with productive interleaving.
+
+**Interleaving Rules**:
+- Avoid clustering 3+ skills from the same modality consecutively
+- Same-modality skills CAN appear in the sequence, just not in long clusters
+- Interleave for rhythm: Exploratory → Evaluative → Generative → Exploratory → Transformative works better than Exploratory → Exploratory → Exploratory → Evaluative
+- Insert `keep-notes` checkpoint every 2-3 skills
+
+**Productive Rhythms**:
+- Discovery: Exploratory → Generative → Evaluative (map → create → filter)
+- Refinement: Evaluative → Transformative → Evaluative (test → reshape → test)
+- Synthesis: Exploratory → Generative → Transformative → Evaluative (map → create → synthesize → select)
+
+**When Clustering IS Appropriate**: Deep investigation may warrant extended Exploratory before switching. Note this explicitly when doing it and explain why.
+
+Present the planned sequence with modality tags and rationale for each skill:
+```
+SKILL SEQUENCE:
+1. [Exploratory] graph-wander - Map territory through associations
+2. [Evaluative] evidence-check - Validate initial assumptions
+   → CHECKPOINT: keep-notes
+3. [Generative] branch - Create distinct solution paths
+4. [Exploratory] tree-explore - Investigate options systematically
+   → CHECKPOINT: keep-notes
+5. [Transformative] integrate - Synthesize findings
+6. [Evaluative] weigh-options - Final selection
+   → CHECKPOINT: keep-notes
+
+Rhythm: E → EV → G → E → T → EV
+```
+
+### Phase 5: Execute
+
+Execute each skill in the planned sequence. At each step:
+
+1. Announce the skill being invoked and its modality
+2. Invoke the skill using `Skill("skill-name")`
+3. At checkpoints, invoke `Skill("introspect-keep-notes")` to persist progress
+
+**Adaptive Re-sequencing**: If discoveries during execution suggest a different path would be more productive, explain the change and modify the sequence. Announce adaptations clearly:
+```
+ADAPTING: Discovery of [what was found] suggests [why change is needed].
+Inserting [skill] before [next planned skill]. Updated sequence: ...
+```
+
+Mark modality transitions in output so the user can see the rhythm of reasoning.
+
+### Phase 6: Checkpoint
+
+After every 2-3 skills, invoke `Skill("introspect-keep-notes")` to:
+- Record key findings from recent skills
+- Assess primitive coverage so far
+- Check if re-sequencing is warranted based on discoveries
+- Maintain context across the session
+
+### Phase 7: Reflect
+
+After executing the sequence:
+
+1. Invoke `Skill("introspect-session-reflect")` to evaluate the reasoning process
+2. Show primitive coverage achieved during the session
+3. Note any adaptations made during execution and what triggered them
+4. Report session artifacts for future reference
+
+```
+REFLECTION:
+Primitive coverage achieved:
+  FOCUS: 85%  MOVE: 70%  GENERATE: 55%  EVALUATE: 90%
+  LINK: 60%   TRANSFORM: 45%  HOLD: 35%
+
+Adaptations made:
+- Added systems-thinking after step 2 (discovered hidden dependencies)
+
+Session artifacts: .thinkies/notes/[session-name].json
+```
+
+## When to Use This Command
+
+Use `/strategize` when you want the reasoning framework itself to be visible and educational. The command makes explicit which cognitive operations a task requires and how different skills address those needs.
+
+Particularly valuable for:
+- Novel problems where you're unsure what kind of thinking is needed
+- Learning to recognize what type of reasoning different situations demand
+- Complex multi-faceted tasks that need coverage across multiple modalities
+- Building intuition for skill selection and composition
+
+For simpler tasks where full primitive analysis isn't needed, use `/reason-session` for structured reasoning or `/reason` for full instrumentation.
+