Cory confirmed the AI alignment agent is Theseus. Reverted all Logos references in skill file, evaluate-trigger, and Alex brief. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
6.6 KiB
Skill: Contribute to Teleo Codex
Ingest source material and extract claims for the shared knowledge base. This skill turns any Claude Code session into a Teleo contributor.
Trigger
/contribute or when the user wants to add source material, extract claims, or propose knowledge to the Teleo Codex.
Prerequisites
- You are running inside a clone of
living-ip/teleo-codex ghCLI is authenticated with access to the repo- User has collaborator access to the repo
Overview
Teleo Codex is a living knowledge base maintained by AI agents and human contributors. You contribute by:
- Archiving source material in
inbox/archive/ - Extracting claims to
domains/{domain}/ - Opening a PR for review by Leo (evaluator) and the domain agent
Step 1: Orient
Read these files to understand the system:
CLAUDE.md— operating rules, schemas, workflowsskills/extract.md— extraction methodologyschemas/source.md— source archive formatschemas/claim.md— claim file format (if it exists)
Identify which domain the contribution targets:
| Domain | Territory | Agent |
|---|---|---|
internet-finance |
domains/internet-finance/ |
Rio |
entertainment |
domains/entertainment/ |
Clay |
ai-alignment |
domains/ai-alignment/ |
Theseus |
health |
domains/health/ |
Vida |
grand-strategy |
core/grand-strategy/ |
Leo |
Step 2: Determine Input Type
Ask the user what they're contributing:
A) URL — Fetch the content, create source archive, extract claims.
B) Text/report — User pastes or provides content directly. Create source archive, extract claims.
C) PDF — User provides a file path. Read it, create source archive, extract claims.
D) Existing source — User points to an unprocessed file already in inbox/archive/. Extract claims from it.
Step 3: Create Branch
git checkout main
git pull origin main
git checkout -b {domain-agent}/contrib-{user}-{brief-slug}
Use the domain agent's name as the branch prefix (e.g., theseus/contrib-alex-alignment-report). This signals whose territory the claims enter.
Step 4: Archive the Source
Create a file in inbox/archive/ following this naming convention:
YYYY-MM-DD-{author-handle}-{brief-slug}.md
Frontmatter template:
---
type: source
title: "Source title"
author: "Author Name"
url: https://original-url-if-exists
date: YYYY-MM-DD
domain: {domain}
format: essay | paper | report | thread | newsletter | whitepaper | news
status: unprocessed
tags: [tag1, tag2, tag3]
contributor: "{user's name}"
---
After the frontmatter, include the FULL content of the source. More content = better extraction.
Step 5: Scan Existing Knowledge
Before extracting, check what already exists to avoid duplicates:
# List existing claims in the target domain
ls domains/{domain}/
# Read titles — each filename IS a claim
# Check for semantic overlap with what you're about to extract
Also scan:
foundations/— domain-independent theorycore/— shared worldview and axioms- The domain agent's beliefs:
agents/{agent}/beliefs.md
Step 6: Extract Claims
Follow skills/extract.md. For each claim:
-
Title IS the claim. Must pass: "This note argues that [title]" works as a sentence.
- Good:
OpenAI's shift to capped-profit created structural misalignment between safety mission and fiduciary obligations.md - Bad:
OpenAI corporate structure.md
- Good:
-
Frontmatter:
---
type: claim
domain: {domain}
description: "one sentence adding context beyond the title"
confidence: proven | likely | experimental | speculative
source: "{contributor name} — based on {source reference}"
created: YYYY-MM-DD
---
- Body:
# [claim title as prose]
[Argument — why this is supported, evidence]
[Inline evidence: cite sources, data, quotes directly in prose]
---
Relevant Notes:
- [[existing-claim-title]] — how it connects
- [[another-claim]] — relationship
Topics:
- [[domain-map]]
-
File location:
domains/{domain}/{slugified-title}.md -
Quality gates (what reviewers check):
- Specific enough to disagree with
- Traceable evidence in the body
- Description adds info beyond the title
- Confidence matches evidence strength
- Not a duplicate of existing claim
- Contradictions are explicit and argued
- Genuinely expands the knowledge base
- All
[[wiki links]]point to real files
Step 7: Update Source Archive
After extraction, update the source file:
status: processed
processed_by: "{contributor name}"
processed_date: YYYY-MM-DD
claims_extracted:
- "claim title 1"
- "claim title 2"
enrichments:
- "existing claim that was enriched"
Step 8: Commit
git add domains/{domain}/*.md inbox/archive/*.md
git commit -m "{agent}/contrib-{user}: add N claims about {topic}
- What: [brief description of claims added]
- Why: [source material, why these matter]
- Connections: [what existing claims these relate to]
Contributor: {user's name}"
The Contributor: trailer is required for human contributions — it ensures attribution.
Step 9: Push and Open PR
git push -u origin {branch-name}
gh pr create \
--title "{agent}/contrib-{user}: {brief description}" \
--body "## Source
{source title and link}
## Claims Proposed
{numbered list of claim titles}
## Why These Matter
{1-2 sentences on value add}
## Contributor
{user's name}
## Cross-Domain Flags
{any connections to other domains the reviewers should check}"
Step 10: What Happens Next
Tell the user:
Your PR is open. Two reviewers will evaluate it:
- Leo — checks quality gates, cross-domain connections, overall coherence
- {Domain agent} — checks domain expertise, duplicates within the domain, technical accuracy
You'll see their feedback as PR comments on GitHub. If they request changes, update your branch and push — they'll re-review automatically.
Your source archive records you as contributor. As claims derived from your work get cited by other claims, your contribution's impact grows through the knowledge graph.
OPSEC
Before committing, verify:
- No dollar amounts, deal terms, or valuations
- No internal business details
- No private communications or confidential information
- When in doubt, ask the user before pushing
Error Handling
- Dirty working tree: Stash or commit existing changes before starting
- Branch conflict: If the branch name exists, append a number or use a different slug
- gh not authenticated: Tell the user to run
gh auth login - Merge conflicts on main:
git pull --rebase origin mainbefore pushing