Swarm Coordination & DOCKET
Relevant source files
The following files were used as context for generating this wiki page:
Swarm Coordination and the DOCKET constitute the operational control plane for the IDAHO-VAULT. This system manages a heterogeneous fleet of AI agents (nodes) collaborating on behalf of a single human principal, Logan Finney. The system architecture enforces strict governance, ensuring that all agentic activity is auditable, traceable, and ultimately gated by human authority.
The DOCKET serves as the live "Pending-Logan" agenda unit, surfacing immediate coordination needs, pending approvals, and active signals. Coordination is facilitated through established protocols like LEVELSET, which provides a persistent shared state and a "source of truth" for all nodes within the swarm.
Sources: LEVELSET 1.md:12-25, coordination.md:9-12, ADJUDICATED.md:200-210
Swarm Architecture & Roles
The swarm is organized into specialized nodes, each with distinct primary and secondary roles. The principal authority resides solely with Logan Finney, who acts as the final arbiter for all high-stakes decisions.
Principal Authority
All swarm activity serves the principal. No node may act autonomously on high-stakes decisions without explicit verification.
Sources: coordination.md:9-12, VAULT-CONVENTIONS.from-pushable-2026-05-08.md:405-415
Swarm Node Registry
| Node | Primary Role | Secondary Role | Platform/Capability |
|---|---|---|---|
| Claude (Anthropic) | Analysis, research, document generation | Framework reasoning | Permanent Authority: CODE |
| Microsoft Copilot | Email coordination, message composition | Institutional lookup | Admin GitHub |
| GitHub Admin Agents | Vault automation, code generation | Swarm orchestration | Automation/CI/CD |
| Linear Agent | Workspace Manager | Decision log keeper | Coordination (Support) |
Sources: coordination.md:14-20, HANDOFF-Swarm-Coordination-2026-03-29.md:46-60
Coordination Hierarchy
The coordination model utilizes a multi-layered structure for task execution and persistence.
The flow of authority originates with the Principal, delegating tasks through Linear, which are then executed as artifacts in the Vault/GitHub.
Sources: HANDOFF-Swarm-Coordination-2026-03-29.md:86-105, VAULT-CONVENTIONS.from-pushable-2026-05-08.md:368-385
The DOCKET System
The DOCKET is the operational "Hot" space (located at !/!/!/) used for live motion. It is specifically designed to surface only items requiring immediate human attention.
DOCKET Components
- Pending Logan: A list of items requiring human eyes, decision, or approval.
- Active Signals/Exhibits: Durable asynchronous agent-to-agent signals or evidence for review.
- Mobile Page: A section for Level 1 Blockers (e.g., Auth failure) requiring immediate human intervention when the principal is AFK.
Sources: ADJUDICATED.md:205-225, PROTOCOL.md:65-75
DOCKET Routing & Lifecycle
Information is not intended to persist in the DOCKET; it must be triaged into permanent records once resolved.
The DOCKET acts as a transient buffer where decisions are staged for human review before being promoted to Vault doctrine or archived.
Sources: ADJUDICATED.md:210-220, VAULT-CONVENTIONS.from-pushable-2026-05-08.md:368-375
Task Handoff Protocols
Handoffs between nodes require strict documentation to prevent "orphaned" processes.
- SENDER: Documents the handoff in
/SWARM/deadletters/if the recipient is unclear and updates theLAST_UPDATED.txttimestamp. - RECEIVER: Acknowledges receipt and updates the
agents.jsonregistry with alast_seentimestamp. - VERIFICATION: The receiver confirms to the principal that the work has been received and is in progress.
Sources: coordination.md:28-34, LEVELSET 1.md:65-75
Conflict Resolution logic
If two nodes receive conflicting instructions:
- Halt all action immediately.
- File a "deadletter" entry describing the conflict.
- Escalate to the principal (Logan) for a resolution.
- Wait for clarification before proceeding.
Sources: coordination.md:36-41, HANDOFF-Swarm-Coordination-2026-03-29.md:110-115
Operational Governance
Governance is anchored in several key files that define the "why," "who," and "what" of the system.
Governance Anchors
| File | Posture | Description |
|---|---|---|
CONSTITUTION.md | Stable | Defines core principles and digital consciousness framing. |
AGENTS.md | Stable | Human-readable registry of capability tiers and bootstrap paths. |
PROTOCOL.md | Workbench | Operational vocabulary and semantics (Hydrate, Ingest, Flag, Handoff). |
DECISIONS.md | Stable | Durable record of confirmed system changes. |
LEVELSET.md | Hot | Consolidated status and synchronization snapshots. |
Sources: PROTOCOL.md:46-52, VAULT-CONVENTIONS.from-pushable-2026-05-08.md:68-85, ADJUDICATED.md:145-160
Swarm Health Indicators
A healthy swarm is measured by the activity of its synchronization files:
agents.jsonupdated within the last 24 hours.- No deadletters older than 48 hours.
- Principal review of
LEVELSETwithin the last week. - All nodes initialized with the current
LEVELSETcontext.
Sources: coordination.md:43-55
Conclusion
Swarm Coordination & DOCKET represents a disciplined approach to managing multi-agent AI environments. By anchoring execution in a human-gated DOCKET and utilizing structured handoff protocols, the system ensures that complex agentic swarms remain subordinate to human journalistic and technical objectives. All state changes are memorialized in the Vault, transforming transient chat-based coordination into a durable record of decisions.
Sources: VAULT-CONVENTIONS.from-pushable-2026-05-08.md:368-372, HANDOFF-Swarm-Coordination-2026-03-29.md:85-90