Claude Integration (The Abhorsen)
Relevant source files
The following files were used as context for generating this wiki page:
Introduction
Claude Integration, identified by the persona The Abhorsen, serves as the primary technical infrastructure and Tier 1 agent within the IDAHO-VAULT ecosystem. Its purpose is to execute directives from the project director, Logan Finney, regarding repository structural work, code execution, and journalism research synthesis. As a Tier 1 agent, Claude possesses direct-write capabilities to the GitHub repository, allowing for atomic commits and infrastructure management.
The integration operates under a constitutional framework defined in !ADMIN/Claude.md, ensuring that all machine-generated content is procedural and clearly distinguished from human-produced Markdown notes. The Abhorsen works alongside other agentic voices like The Lexicographer (Codex) and The Clerk (Copilot) to maintain the vault's integrity.
Sources: LEVELSET-v2.md:12, !/!/AUDIT-AGENTIC-VOICES-2026-04-03.md:16, HANDOFF-ADMIN-2026-03-15.md:28
Architecture and Agent Tiers
The IDAHO-VAULT utilizes a tiered capability model to manage agent access and responsibilities. Claude Integration occupies the most privileged tier for technical execution.
Agent Tiers
| Tier | Designation | Capability | Role |
|---|---|---|---|
| Tier 1 | Direct Write | Full repo write access via Claude Code CLI. | Structural work, Python execution, PR management. |
| Tier 2 | Constitutional | Constitutional layer and judgment calls. | Governance, conventions, and policy enforcement. |
| Tier 3 | Read/Analysis | Read-only access to repository context. | Synthesis and reporting (e.g., Levelset synthesis). |
Sources: LEVELSET-v2.md:79-92, !/!/AUDIT-AGENTIC-VOICES-2026-04-03.md:16-30, LEVELSET-v3.2.6.1-PROMPT.md:111-118
The Control Plane
The Abhorsen is anchored by a specific configuration layer that loads automatically upon session initialization.
The diagram shows the initialization flow where the Claude Code CLI loads repository-specific instructions and settings to establish the agent's identity and operational boundaries.
Sources: !/!/AUDIT-AGENTIC-VOICES-2026-04-03.md:65, sun_mar_15_2026_admin_git_hub_copilot.json:280
Operational Protocols
Claude Integration follows strict protocols to ensure transparency and prevent "hallucinated governance" or unauthorized changes.
LEVELSET Protocol
LEVELSET is a permanent, auditable checkpoint protocol used to synchronize context across multiple Claude conversations. It ensures that session memory is bridged with long-term architectural goals.
- Additive Only: Reports are versioned files in
!ADMIN/that are never deleted or overwritten. - Source of Truth: It synthesizes reports into a canonical record for verification by the Director.
- Signals: Claude uses signals like
ESCALATE,BLOCK,COLLISION, andDISCOVERYto push urgent flags between cycles.
Sources: LEVELSET-v2.md:12-20, LEVELSET-v3.2.6.1-PROMPT.md:144-150, Sheet20.md:38-42
Security and Collaboration
Claude collaborates with Codex (The Lexicographer) for security audits and branch protection. While Claude handles implementation (e.g., fixing CI workflows, standardizing actions), Codex performs reliability audits on clean pushes to identify "flapping" checks.
The sequence illustrates the shared security workflow where Logan remains the sole decision authority for branch protection rules.
Sources: PROPOSAL-CLAUDE-CODEX-SECURITY-COLLAB-2026-05-25.md:52-85
Key Artifacts and Configuration
Claude's behavior is governed by several critical files within the repository structure.
Configuration Registry
.claude/CLAUDE.md: Official CLI path for repo-wide instructions, including allowed tools and 1Password requirements.!ADMIN/Claude.md: The primary "constitution" defining inter-agent communication and the DHPP (Digital Hygiene & Privacy Protocol).swarm.json: Machine-readable registry of agents, personas, and protocols.
Sources: !/!/AUDIT-AGENTIC-VOICES-2026-04-03.md:65-79, Sheet20.md:196
Digital Hygiene & Privacy Protocol (DHPP)
Formerly known as the "Wall of Secrecy," the DHPP is a technical procedure for compartmentalizing sensitive information.
- Off-the-Record (OTR) Data: Routed exclusively to the
!ADMIN/directory. - Public Exclusion: OTR content is barred from the public repository and excluded from general model training to prevent data bleeding.
Sources: Sheet20.md:200-205, LEVELSET-v3.2.6.1-PROMPT.md:38-42
Technical Execution (Tier 1)
As a Tier 1 agent, Claude is responsible for procedural machine products, primarily written in Python.
Deployed Automations
| Script | Purpose | Status |
|---|---|---|
sort_audit.py | Audits vault organization and flags classification errors. | Deployed to main. |
propose_moves.py | Proposes automated file moves for vault cleanup. | Deployed to main. |
idaho_leg_scraper.py | Scrapes legislative data from official Idaho sources. | Active on feature branch. |
wayback_audit.py | Checks for dead links and initiates archival via Wayback Machine. | Deployed to main. |
Sources: LEVELSET-v2.md:40-50, HANDOFF-ADMIN-2026-03-15.md:15-20
Git Fingerprinting
Claude identifies itself in the git history through specific signatures:
- Identity:
Claude(The Abhorsen). - Branch Prefix:
claude/. - Commit Suffix:
-C. - Labels:
agent:claude-code.
Sources: !/!/AUDIT-AGENTIC-VOICES-2026-04-03.md:48-55
Summary
Claude Integration (The Abhorsen) provides the IDAHO-VAULT with a robust execution layer for repository maintenance and research automation. By operating under the LEVELSET protocol and the DHPP, it maintains a strict boundary between public journalism artifacts and private administrative governance. Its role as a Tier 1 direct-write agent allows for efficient, atomic updates to project infrastructure while remaining strictly subordinate to the human director's authority.
Sources: LEVELSET-v2.md:170-180, Sheet20.md:126