/**
* Session Analysis Guide Resource Template
*
* Provides a process guide for qualitative session analysis.
* Accessed via resource template: thoughtbox://session-analysis-guide
*/
/**
* Session analysis process guide content
*
* This guide teaches the calling agent how to perform qualitative analysis
* on a reasoning session. The server provides objective structural metrics;
* this guide helps the client identify key moments and extract learnings.
*/
export const SESSION_ANALYSIS_GUIDE = `# Session Analysis Process
You have received structural metrics for a reasoning session. To perform qualitative analysis and extract learnings, follow this process:
## Step 1: Review Session Content
Use the \`session\` tool with operation \`get\` to retrieve the full session with all thoughts if you haven't already.
> **Note**: If tools appear unavailable, use \`thoughtbox_gateway\` with operation \`session\` instead. The gateway is always enabled and routes to the session handler internally.
## Step 2: Identify Key Moments
Scan through the thoughts looking for:
1. **Pivots**: Where did reasoning direction change significantly?
2. **Decisions**: Which thoughts resolved uncertainty or chose between options?
3. **Insights**: Where did synthesis or "aha moments" occur?
4. **Revisions**: What corrections were made? (Check \`isRevision\` field)
5. **Branch Points**: Where did exploration diverge? (Check \`branchFromThought\` field)
## Step 3: Assess Significance
For each key moment identified, consider:
- **Impact**: How much did this thought influence the outcome?
- **Novelty**: Was this a new approach or standard reasoning?
- **Transferability**: Could this pattern apply to other problems?
## Step 4: Extract Learnings
Use the \`session\` tool with operation \`extract_learnings\` and your identified key moments:
\`\`\`json
{
"sessionId": "<session-id>",
"keyMoments": [
{
"thoughtNumber": 5,
"type": "decision",
"significance": 8,
"summary": "Chose hybrid approach over pure options"
}
],
"targetTypes": ["pattern", "anti-pattern", "signal"]
}
\`\`\`
## Step 5: Generate Artifacts
The extract_learnings tool will generate:
- **Patterns**: From decisions and insights (for DGM experiments)
- **Anti-patterns**: From revisions (what didn't work)
- **Signals**: Session-level success/failure metrics
## Quality Indicators in Metrics
Use the structural metrics to guide your analysis:
| Metric | Interpretation |
|--------|----------------|
| High \`revisionRate\` | Look for anti-patterns - what needed correction |
| Low \`linearityScore\` | Look for branch exploration patterns |
| \`critiqueRequests\` > 0 | Check critique responses for quality signals |
| \`isComplete\` = true | Session reached natural conclusion |
| \`hasConvergence\` = true | Branching resolved to a decision |
## Key Moment Detection Heuristics
When scanning thoughts, look for these signals:
### Pivot Indicators
- Phrases like "actually", "wait", "on second thought"
- Significant shift in topic or approach
- Abandoning one line of reasoning for another
### Decision Indicators
- Explicit choices: "I'll go with", "choosing", "decided"
- Comparison conclusions: "X is better because"
- Commitment statements
### Insight Indicators
- Synthesis: "This means that", "so the pattern is"
- Connections: "This is related to", "similar to"
- Realizations: "I see now", "the key insight"
## Writing Effective Learnings
When providing key moments to \`extract_learnings\`:
1. **Be specific**: Include the exact thought number
2. **Rate significance**: 1-10 scale helps prioritize
3. **Summarize why**: Brief explanation aids future retrieval
4. **Choose correct type**: decision, pivot, insight, revision, or branch
## Evolution Integration
Extracted learnings feed into the DGM evolution system:
- **Patterns** become experimental entries in \`.claude/rules/evolution/experiments/\`
- **Anti-patterns** document what to avoid
- **Signals** append to \`.claude/rules/evolution/signals.jsonl\`
Over time, successful patterns gain fitness and may be promoted to main rules.
> **Note on paths**: The \`.claude/rules/evolution/\` paths refer to the Claude Code memory system directory structure, not Thoughtbox internal paths. These are filesystem paths where the calling Claude Code agent stores learnings for its own memory system. Thoughtbox extracts the learnings; the agent decides where to persist them.
---
*This guide is available via the thoughtbox://session-analysis-guide resource.*
`;
/**
* Returns the resource template definition for session analysis guide
* Used by the ListResourceTemplatesRequestSchema handler
*/
export function getSessionAnalysisResourceTemplates() {
return {
resourceTemplates: [
{
uriTemplate: "thoughtbox://session-analysis-guide",
name: "session-analysis-guide",
title: "Session Analysis Process Guide",
description:
"Process guide for qualitative analysis of reasoning sessions. Teaches how to identify key moments and extract learnings for the DGM evolution system.",
mimeType: "text/markdown",
annotations: {
audience: ["assistant"],
priority: 0.8,
},
},
],
};
}
/**
* Resolves a thoughtbox://session-analysis-guide URI to its content
*
* @param uri - The resource URI to resolve
* @returns Resource content with metadata
*/
export function getSessionAnalysisGuideContent(uri: string) {
if (uri !== "thoughtbox://session-analysis-guide") {
throw new Error(
`Invalid session analysis guide URI: ${uri}. Expected: thoughtbox://session-analysis-guide`
);
}
return {
uri,
mimeType: "text/markdown",
text: SESSION_ANALYSIS_GUIDE,
};
}
