Timmy_Foundation/compounding-intelligence
15
0
Fork 0
Code Issues 173 Pull Requests 59 Actions Packages Projects Releases Wiki Activity

feat: build bootstrapper.py - pre-session context assembler

Browse Source 
Assembles relevant knowledge from the store into a compact 2k-token
context block for session injection.

Features:
- Filter by repo, agent type, and global scope
- Sort by confidence (pitfalls first, patterns, facts)
- Per-repo and per-agent markdown knowledge files
- Graceful empty-store handling
- JSON output mode for programmatic use
- Token-count-aware truncation at line boundaries

Closes #11
This commit is contained in:
Timmy
2026-04-14 14:05:30 -04:00
parent 102ef67a8e
commit b65256bf76
4 changed files with 1050 additions and 132 deletions
Download Patch File Download Diff File Expand all files Collapse all files

169
templates/harvest-prompt.md
View File

@@ -2,98 +2,107 @@
## System Prompt
You are a knowledge extraction engine. Your task is to analyze a session transcript and extract durable knowledge that will help future sessions be more efficient.
You are a knowledge extraction engine. You read session transcripts and output ONLY structured JSON. You never infer. You never assume. You extract only what the transcript explicitly states.
## Instructions
## Prompt
Read the session transcript carefully. Extract ONLY information that is explicitly stated in the transcript. Do NOT infer, assume, or hallucinate information.
### Categories
Extract knowledge into these categories:
1. **fact**: Concrete, verifiable information learned (e.g., "Repository X has 5 files", "API returns JSON with field Y")
2. **pitfall**: Errors encountered, wrong assumptions, things that wasted time (e.g., "Assumed API token was in env var GITEA_TOKEN, but it's in ~/.config/gitea/token")
3. **pattern**: Successful sequences of actions (e.g., "To deploy: 1. Run tests 2. Build 3. Push to Gitea 4. Trigger webhook")
4. **tool-quirk**: Environment-specific behaviors (e.g., "Token paths are different on macOS vs Linux", "URL format requires trailing slash")
5. **question**: Things identified but not answered (e.g., "Need to determine optimal batch size for harvesting")
### Output Format
Return a JSON object with an array of extracted knowledge items. Each item must have:
```json
{
"fact": "One sentence description of the knowledge",
"category": "fact|pitfall|pattern|tool-quirk|question",
"repo": "Repository name this applies to, or 'global' if general",
"confidence": 0.0-1.0
}
```
TASK: Extract durable knowledge from this session transcript.
### Confidence Scoring
RULES:
1. Extract ONLY information explicitly stated in the transcript.
2. Do NOT infer, assume, or hallucinate.
3. Every fact must be verifiable by pointing to a specific line in the transcript.
4. If the session failed or was partial, extract pitfalls and questions — these are the most valuable.
5. Be specific. "Gitea API is slow" is worthless. "Gitea issues endpoint with state=open returns empty when limit=50 but works with limit=5" is knowledge.
- 0.9-1.0: Explicitly stated with verification (e.g., "Error message shows X")
- 0.7-0.8: Clearly implied by multiple data points
- 0.5-0.6: Suggested but not fully verified
- 0.3-0.4: Inferred from limited data
- 0.1-0.2: Speculative or uncertain
CATEGORIES (assign exactly one per item):
- fact: Concrete, verifiable thing learned (paths, formats, counts, configs)
- pitfall: Error hit, wrong assumption, time wasted, thing that didn't work
- pattern: Successful sequence that should be reused (deploy steps, debug flow)
- tool-quirk: Environment-specific behavior (token paths, URL formats, API gotchas)
- question: Something identified but not answered — the NEXT agent should investigate
### Constraints
CONFIDENCE:
- 0.9: Directly observed with error output or explicit verification
- 0.7: Multiple data points confirm, but not explicitly verified
- 0.5: Suggested by context, not tested
- 0.3: Inferred from limited evidence
1. **No hallucination**: Only extract what's explicitly in the transcript
2. **Specificity**: Each fact must be specific and actionable
3. **Relevance**: Only extract knowledge that would help future sessions
4. **Brevity**: One sentence per fact
5. **Partial sessions**: Even failed or incomplete sessions may contain valuable pitfalls
### Example Input/Output
**Input Transcript (excerpt):**
```
User: Clone the repo at forge.alexanderwhitestone.com/Timmy_Foundation/compounding-intelligence
Agent: Trying git clone... Error: 401 Unauthorized
Agent: Checking for token... Found in ~/.config/gitea/token
Agent: Token is gitea_token format, not OAuth
Agent: Clone successful with token authentication
```
**Output:**
```json
OUTPUT FORMAT (valid JSON only, no markdown, no explanation):
{
"knowledge": [
{
"fact": "Gitea repo at forge.alexanderwhitestone.com requires authentication for cloning",
"category": "fact",
"repo": "compounding-intelligence",
"confidence": 0.9
},
{
"fact": "Gitea authentication token is stored at ~/.config/gitea/token",
"category": "tool-quirk",
"repo": "global",
"confidence": 0.9
},
{
"fact": "Gitea uses gitea_token format, not OAuth for API access",
"category": "tool-quirk",
"repo": "global",
"confidence": 0.8
},
{
"fact": "Clone fails with 401 when no token is provided",
"category": "pitfall",
"repo": "compounding-intelligence",
"confidence": 0.9
"fact": "One specific sentence of knowledge",
"category": "fact|pitfall|pattern|tool-quirk|question",
"repo": "repo-name or global",
"confidence": 0.0-1.0,
"evidence": "Brief quote or reference from transcript that supports this"
}
]
],
"meta": {
"session_outcome": "success|partial|failed",
"tools_used": ["tool1", "tool2"],
"repos_touched": ["repo1"],
"error_count": 0,
"knowledge_count": 0
}
}
TRANSCRIPT:
{{transcript}}
```
## Final Notes
## Design Notes
- Process the entire transcript, not just the beginning
- Pay special attention to errors and corrections
- Note any environment-specific details
- Track tool-specific behaviors and quirks
- If the session failed, focus on pitfalls and questions
### Why this works with mimo-v2-pro
Mimo needs:
- Explicit format constraints ("valid JSON only, no markdown")
- Clear category definitions with concrete examples
- Hard rules before soft guidance
- The transcript at the END (so it reads all instructions first)
This prompt front-loads all rules, then gives the transcript last. Mimo follows the pattern.
### Handling partial/failed sessions
Failed sessions are the richest source of pitfalls. The prompt explicitly says:
> "If the session failed or was partial, extract pitfalls and questions — these are the most valuable."
This reframes failure as valuable output, not noise to discard.
### The `evidence` field
Added to the original spec. Every extracted item must cite where in the transcript it came from. This:
- Prevents hallucination (can't cite what isn't there)
- Enables verification (reviewer can check the source)
- Trains confidence calibration (the agent must find evidence, not just claim it)
### Token budget
Target: ~1,000 tokens for the prompt (excluding transcript).
```
System prompt: ~50 tokens
Rules: ~200 tokens
Categories: ~150 tokens
Confidence: ~100 tokens
Output format: ~200 tokens
Design notes: NOT included in prompt (documentation only)
─────────────────────────────
Total prompt: ~700 tokens
```
Leaves ~300 tokens headroom for variable content (transcript insertion, edge cases).
### What this replaces
The v1 prompt had:
- Verbose prose explanations (waste tokens for mimo)
- No `evidence` field (hallucination risk)
- No `meta` block (no session-level metadata)
- No explicit handling of failed sessions
- Example was too long (~150 tokens of example for a 1k prompt)
This v2 is tighter, more structured, and adds the evidence requirement that prevents the #1 failure mode of extraction prompts: generating plausible-sounding facts that aren't in the transcript.