Three-agent knowledge base (Leo, Rio, Clay) with: - 177 claim files across core/ and foundations/ - 38 domain claims in internet-finance/ - 22 domain claims in entertainment/ - Agent soul documents (identity, beliefs, reasoning, skills) - 14 positions across 3 agents - Claim/belief/position schemas - 6 shared skills - Agent-facing CLAUDE.md operating manual Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
75 lines
2.5 KiB
Markdown
75 lines
2.5 KiB
Markdown
# Skill: Extract
|
|
|
|
Turn raw content into structured evidence and proposed claims.
|
|
|
|
## When to Use
|
|
|
|
When new content arrives in inbox/ — articles, tweets, papers, transcripts, research files.
|
|
|
|
## Input
|
|
|
|
Raw content (text, URL, document).
|
|
|
|
## Process
|
|
|
|
### Step 1: Read the source material completely
|
|
|
|
Don't skim. Read the full content before extracting anything. Understand the author's argument, not just individual data points.
|
|
|
|
### Step 2: Separate evidence from interpretation
|
|
|
|
**Evidence** is factual: data, statistics, quotes, study results, events, observations. Things that are verifiable regardless of your interpretive framework.
|
|
|
|
**Claims** are interpretive: assertions about what the evidence means, causal relationships, predictions, evaluations. Things someone could disagree with.
|
|
|
|
Most sources mix these freely. Your job is to separate them.
|
|
|
|
### Step 3: Extract evidence
|
|
|
|
For each piece of evidence:
|
|
- Is it sourced and verifiable?
|
|
- Is it relevant to at least one Teleo domain?
|
|
- Does it already exist in the knowledge base? (check for duplicates)
|
|
|
|
Include evidence inline in the claim body — cite sources, data, studies directly in the prose.
|
|
|
|
### Step 4: Extract candidate claims
|
|
|
|
For each potential claim:
|
|
- Is it specific enough to disagree with? ("AI is changing healthcare" → NO. "AI diagnostic triage achieves 97% sensitivity across 14 conditions" → YES)
|
|
- Does it cite evidence from this source or the knowledge base?
|
|
- Does it duplicate an existing claim? (semantic check — different words, same idea)
|
|
- Title passes the prose-as-claim test: "This note argues that [title]" works as a sentence
|
|
|
|
Create candidate claim files for evaluation.
|
|
|
|
### Step 5: Classify by domain
|
|
|
|
Tag each evidence and claim with primary domain:
|
|
- internet-finance, entertainment, grand-strategy
|
|
|
|
Cross-domain items get a primary domain + secondary_domains list.
|
|
|
|
### Step 6: Identify enrichments
|
|
|
|
Does this source contain information that would improve existing notes?
|
|
- New data for existing claims
|
|
- Counter-evidence to existing claims
|
|
- New connections between existing claims
|
|
|
|
Flag enrichments for the evaluation cycle.
|
|
|
|
## Output
|
|
|
|
- Claim files in domains/{domain}/ with evidence inline
|
|
- Candidate claim files for PR review
|
|
- Enrichment flags for existing notes
|
|
- Extraction summary: N evidence extracted, N claims proposed, N enrichments flagged
|
|
|
|
## Quality Gate
|
|
|
|
- Every claim cites verifiable evidence inline
|
|
- Every claim is specific enough to disagree with
|
|
- No duplicates of existing knowledge base content
|
|
- Domain classification is accurate
|
|
- Titles work as prose propositions
|