Name: search-tree
Author: Yugoge

Prompt file imported from Yugoge/claude-brain (.claude/commands/search-tree.md). Fill in {{arguments}} before use. Copyright stays with the author.

Explore multiple solution paths for: {{arguments}}

Methodology: Tree Search with Path Evaluation

Inspired by MCTS (Monte Carlo Tree Search) and LATS (Language Agent Tree Search).

Phase 1: Path Generation

Generate 3-5 distinct possible search paths to answer the question.

Example: Question "How to start an AI company in 2025?" Possible paths:

Path A: Technical approach (infrastructure, tools, ML stack)
Path B: Business approach (funding, market, customers)
Path C: Legal approach (incorporation, regulations, IP)
Path D: Talent approach (hiring, team building, culture)
Path E: Product approach (MVP, validation, iteration)

Phase 2: Initial Path Exploration (parallel)

For each path, execute 1-2 exploratory searches in parallel.

Prompt for each:

Explore the "{{arguments}}" question from the [PATH NAME] angle.
What are the key considerations, steps, or information needed?
Provide: main points, potential challenges, resources needed.

Phase 3: Path Evaluation

For each path, score based on:

Relevance (0-3): How directly it addresses the question
Completeness (0-3): How fully it answers the question
Actionability (0-3): How practical/implementable the info is
Evidence Quality (0-1): Quality of sources found

Total score: 0-10 per path

Phase 4: Deep Dive on Top Paths

Select the top 2 highest-scoring paths. For each, execute 3-5 deep searches:

Deep dive into [PATH NAME] for "{{arguments}}".
Find: specific steps, expert advice, case studies, data.
Prioritize: actionable information, recent examples, proven methods.

Execute these in parallel.

Phase 5: Recursive Refinement (optional)

If the top path reveals new sub-paths worth exploring:

Generate 2-3 sub-paths within the best path
Repeat evaluation and exploration (max depth: 2 levels)

Phase 6: Integration & Decision

Synthesize findings from all explored paths:

## Tree Search Report: {{arguments}}

### Question
{{arguments}}

### Paths Explored
1. **Path A: [Name]** - Score: [X/10]
   - Key findings: [...]
   - Strengths: [...]
   - Limitations: [...]

2. **Path B: [Name]** - Score: [X/10]
   - Key findings: [...]
   - Strengths: [...]
   - Limitations: [...]

[Continue for all paths]

### Recommended Path(s)
Based on evaluation, the optimal approach is: **[Path Name(s)]**

Reasoning: [Why this path scored highest]

### Integrated Solution
[Combine insights from multiple paths into cohesive answer]

### Implementation Steps
1. [Actionable step from best path]
2. [...]
3. [Consider incorporating elements from secondary paths]

### Alternative Approaches
[Brief summary of other viable paths not chosen]

### Decision Tree Visualization

Question: {{arguments}} ├─ Path A [Score: X/10] → [Outcome] ├─ Path B [Score: X/10] → [Outcome] ✓ SELECTED │ ├─ Sub-path B1 → [...] │ └─ Sub-path B2 → [...] ├─ Path C [Score: X/10] → [Outcome] └─ Path D [Score: X/10] → [Outcome]


### Sources by Path
**Path A**: [URLs]
**Path B**: [URLs]
...

### Reflection
What worked: [...]
What didn't: [...]
If I were to search again: [...]

Execution Guidelines

Diverse paths: Ensure paths approach from different angles
Parallel exploration: Explore all paths simultaneously initially
Honest scoring: Don't force a path to work if evidence is weak
Prune dead ends: If a path scores <4, don't deep dive
Combine insights: Best solution often integrates multiple paths
Use TodoWrite: Track which paths are being explored

Best For

Open-ended questions with multiple valid approaches
Problems requiring evaluation of trade-offs
Strategic decisions with uncertainty
Research where the "best" path isn't obvious

Install search-tree and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

Explore multiple solution paths for: **{{arguments}}** ## Methodology: Tree Search with Path Evaluation Inspired by MCTS (Monte Carlo Tree Search) and LATS (Language Agent Tree Search). ### Phase 1: Path Generation Generate 3-5 distinct possible search paths to answer the question. Example: Question "How to start an AI company in 2025?" Possible paths: - Path A: Technical approach (infrastructure, tools, ML stack) - Path B: Business approach (funding, market, customers) - Path C: Legal approach (incorporation, regulations, IP) - Path D: Talent approach (hiring, team building, culture) - Path E: Product approach (MVP, validation, iteration) ### Phase 2: Initial Path Exploration (parallel) For each path, execute 1-2 exploratory searches **in parallel**. Prompt for each: ``` Explore the "{{arguments}}" question from the [PATH NAME] angle. What are the key considerations, steps, or information needed? Provide: main points, potential challenges, resources needed. ``` ### Phase 3: Path Evaluation For each path, score based on: - **Relevance** (0-3): How directly it addresses the question - **Completeness** (0-3): How fully it answers the question - **Actionability** (0-3): How practical/implementable the info is - **Evidence Quality** (0-1): Quality of sources found Total score: 0-10 per path ### Phase 4: Deep Dive on Top Paths Select the top 2 highest-scoring paths. For each, execute 3-5 deep searches: ``` Deep dive into [PATH NAME] for "{{arguments}}". Find: specific steps, expert advice, case studies, data. Prioritize: actionable information, recent examples, proven methods. ``` Execute these **in parallel**. ### Phase 5: Recursive Refinement (optional) If the top path reveals new sub-paths worth exploring: - Generate 2-3 sub-paths within the best path - Repeat evaluation and exploration (max depth: 2 levels) ### Phase 6: Integration & Decision Synthesize findings from all explored paths: ```markdown ## Tree Search Report: {{arguments}} ### Question {{arguments}} ### Paths Explored 1. **Path A: [Name]** - Score: [X/10] - Key findings: [...] - Strengths: [...] - Limitations: [...] 2. **Path B: [Name]** - Score: [X/10] - Key findings: [...] - Strengths: [...] - Limitations: [...] [Continue for all paths] ### Recommended Path(s) Based on evaluation, the optimal approach is: **[Path Name(s)]** Reasoning: [Why this path scored highest] ### Integrated Solution [Combine insights from multiple paths into cohesive answer] ### Implementation Steps 1. [Actionable step from best path] 2. [...] 3. [Consider incorporating elements from secondary paths] ### Alternative Approaches [Brief summary of other viable paths not chosen] ### Decision Tree Visualization ``` Question: {{arguments}} ├─ Path A [Score: X/10] → [Outcome] ├─ Path B [Score: X/10] → [Outcome] ✓ SELECTED │ ├─ Sub-path B1 → [...] │ └─ Sub-path B2 → [...] ├─ Path C [Score: X/10] → [Outcome] └─ Path D [Score: X/10] → [Outcome] ``` ### Sources by Path **Path A**: [URLs] **Path B**: [URLs] ... ### Reflection What worked: [...] What didn't: [...] If I were to search again: [...] ``` ## Execution Guidelines 1. **Diverse paths**: Ensure paths approach from different angles 2. **Parallel exploration**: Explore all paths simultaneously initially 3. **Honest scoring**: Don't force a path to work if evidence is weak 4. **Prune dead ends**: If a path scores <4, don't deep dive 5. **Combine insights**: Best solution often integrates multiple paths 6. **Use TodoWrite**: Track which paths are being explored ## Best For - Open-ended questions with multiple valid approaches - Problems requiring evaluation of trade-offs - Strategic decisions with uncertainty - Research where the "best" path isn't obvious

{ "prompt": "Explore multiple solution paths for: **{{arguments}}**\n\n## Methodology: Tree Search with Path Evaluation\n\nInspired by MCTS (Monte Carlo Tree Search) and LATS (Language Agent Tree Search).\n\n### Phase 1: Path Generation\nGenerate 3-5 distinct possible search paths to answer the question.\n\nExample: Question \"How to start an AI company in 2025?\"\nPossible paths:\n- Path A: Technical approach (infrastructure, tools, ML stack)\n- Path B: Business approach (funding, market, customers)\n- Path C: Legal approach (incorporation, regulations, IP)\n- Path D: Talent approach (hiring, team building, culture)\n- Path E: Product approach (MVP, validation, iteration)\n\n### Phase 2: Initial Path Exploration (parallel)\nFor each path, execute 1-2 exploratory searches **in parallel**.\n\nPrompt for each:\n```\nExplore the \"{{arguments}}\" question from the [PATH NAME] angle.\nWhat are the key considerations, steps, or information needed?\nProvide: main points, potential challenges, resources needed.\n```\n\n### Phase 3: Path Evaluation\nFor each path, score based on:\n- **Relevance** (0-3): How directly it addresses the question\n- **Completeness** (0-3): How fully it answers the question\n- **Actionability** (0-3): How practical/implementable the info is\n- **Evidence Quality** (0-1): Quality of sources found\n\nTotal score: 0-10 per path\n\n### Phase 4: Deep Dive on Top Paths\nSelect the top 2 highest-scoring paths.\nFor each, execute 3-5 deep searches:\n```\nDeep dive into [PATH NAME] for \"{{arguments}}\".\nFind: specific steps, expert advice, case studies, data.\nPrioritize: actionable information, recent examples, proven methods.\n```\n\nExecute these **in parallel**.\n\n### Phase 5: Recursive Refinement (optional)\nIf the top path reveals new sub-paths worth exploring:\n- Generate 2-3 sub-paths within the best path\n- Repeat evaluation and exploration (max depth: 2 levels)\n\n### Phase 6: Integration & Decision\nSynthesize findings from all explored paths:\n\n```markdown\n## Tree Search Report: {{arguments}}\n\n### Question\n{{arguments}}\n\n### Paths Explored\n1. **Path A: [Name]** - Score: [X/10]\n - Key findings: [...]\n - Strengths: [...]\n - Limitations: [...]\n\n2. **Path B: [Name]** - Score: [X/10]\n - Key findings: [...]\n - Strengths: [...]\n - Limitations: [...]\n\n[Continue for all paths]\n\n### Recommended Path(s)\nBased on evaluation, the optimal approach is: **[Path Name(s)]**\n\nReasoning: [Why this path scored highest]\n\n### Integrated Solution\n[Combine insights from multiple paths into cohesive answer]\n\n### Implementation Steps\n1. [Actionable step from best path]\n2. [...]\n3. [Consider incorporating elements from secondary paths]\n\n### Alternative Approaches\n[Brief summary of other viable paths not chosen]\n\n### Decision Tree Visualization\n```\nQuestion: {{arguments}}\n├─ Path A [Score: X/10] → [Outcome]\n├─ Path B [Score: X/10] → [Outcome] ✓ SELECTED\n│ ├─ Sub-path B1 → [...]\n│ └─ Sub-path B2 → [...]\n├─ Path C [Score: X/10] → [Outcome]\n└─ Path D [Score: X/10] → [Outcome]\n```\n\n### Sources by Path\n**Path A**: [URLs]\n**Path B**: [URLs]\n...\n\n### Reflection\nWhat worked: [...]\nWhat didn't: [...]\nIf I were to search again: [...]\n```\n\n## Execution Guidelines\n1. **Diverse paths**: Ensure paths approach from different angles\n2. **Parallel exploration**: Explore all paths simultaneously initially\n3. **Honest scoring**: Don't force a path to work if evidence is weak\n4. **Prune dead ends**: If a path scores <4, don't deep dive\n5. **Combine insights**: Best solution often integrates multiple paths\n6. **Use TodoWrite**: Track which paths are being explored\n\n## Best For\n- Open-ended questions with multiple valid approaches\n- Problems requiring evaluation of trade-offs\n- Strategic decisions with uncertainty\n- Research where the \"best\" path isn't obvious", "variables": [ "arguments" ], "notes": "Tree search exploration with MCTS-inspired path evaluation Arguments: <question-or-problem>.", "metadata": { "source": { "provider": "github-claude-commands", "repository": "https://github.com/Yugoge/claude-brain", "path": ".claude/commands/search-tree.md", "ref": "9ef00bb7c576c10ae605666be6d9c1860c330b44", "url": "https://github.com/Yugoge/claude-brain/blob/9ef00bb7c576c10ae605666be6d9c1860c330b44/.claude/commands/search-tree.md", "key": "Yugoge/claude-brain/.claude/commands/search-tree.md" } } }

search-tree

About