Skip to content

Current Visual Guide

Lucas YourLastName edited this page Oct 31, 2026 · 5 revisions

Current visual guide and agent roles

These four guides extend the foundational four-part field guide. Seven concise role cards and seven detailed agent sheets describe the registered agent roles. The diagrams explain file-backed responsibilities; they do not create processes or grant permissions. Read the current manuals before running a command.

Each image is a standalone SVG with selectable text, a title and description, and one embedded generated art background. The 7680 × 4320 SVG canvas and matching 8K raster exports provide crisp vector labels. The first-set background art was generated at 1672 × 941; the detailed sheet backgrounds were generated at 1672 × 941 or 1536 × 1024. Scaling these backgrounds does not add native illustration detail. Asset provenance records exact hashes and dimensions. The art uses glass panels and role accents, while every step and boundary is also written as text. Color alone does not carry meaning.

Published beta images. The tagged source contains 18 beta SVGs and four retained alpha.3 4K JPEGs. The 18 separate 8K JPEG exports are release assets. The first eleven exports are below; the seven detailed agent field guides link the remaining seven. All 18 release JPEG downloads matched their recorded bytes and SHA-256 after authenticated readback. Full checksums.

8K release export Bytes SHA-256
systemx-agent-0-8k.jpg 2,821,967 65fdc8cfa9324f8969ebb046b93d880318eceab51927423522046435a2e9bf17
systemx-agent-1-8k.jpg 2,897,396 fc02c7a7618bdc09119db427a146582d91ef98cb77d43d00efbafe5ccba72788
systemx-agent-alpha-8k.jpg 3,506,938 1ecb1348150bdcf3d21e3da200d207f32fb28bf09cb9220e759a8646f89eaaf8
systemx-agent-beta-8k.jpg 2,725,562 6e8203ae43a888d3ec4edd7db2f59c6014573478d0d61774c6890c83c997b609
systemx-agent-x-8k.jpg 2,772,452 26eb1be283b2f777b68efb14ab6162af6f6fd61a388404c0ae9f144853bde505
systemx-agent-y-8k.jpg 2,778,809 9424c299f061e950bc560e7ef968afac884279ae7cc18a03347af623405a53fe
systemx-agent-z-8k.jpg 2,868,544 aeec87d65a76c320af2777552441d707fe1bbc6324d6a1af8b895aa89dbfd1a8
systemx-guide-access-matrix-8k.jpg 2,790,777 a5c5b934932f200c56bf533ca8383600add358a4e5bf2ea638f376122b90ecff
systemx-guide-bot-workspaces-8k.jpg 2,830,948 8c7faef5222137a12c5a8c4a45ab533a1e4e1273c81b5232513084a5687bbee5
systemx-guide-lifecycle-8k.jpg 3,024,819 10a73904adaf6c77386c869d8ea17393b8c3085ac850a9d3521a5b5818b39afd
systemx-guide-security-access-8k.jpg 3,138,168 bd06f9039d5cdb56d2468a824ab8a803faa2d99919d15a2897db65577578ff90

Four current guides

Project lifecycle

Six panels for setup, scope selection, role activation, review, release, and recovery.

Full-size SVG. An operator first selects a project and installs reviewed blank defaults, then activates only the roles needed for the task. Agent 0 keeps canonical work and acceptance in WORK/TASKS.json. A release tag identifies reviewed source, while a deployment receipt records a separate target action. Update and restore use retained receipts. A folder or role record does not start a worker or deploy a service.

Named BOT and DOT workspaces

A canonical task points to a named BOT or DOT workspace; a broken connector separates that record from an external account.

Full-size SVG. Agent Beta records one named workspace under AGENTB/.SYSTEMXB-BOT_NAME, including a reviewed identity and scope. Canonical TASK references remain owned by Agent 0. An external provider account is a separate authority. Registration does not authenticate the bot, start a provider session, or grant an account permission. Agent Y continues to own an explicitly selected CLI handoff.

Security and access

A review checklist leads to protected key entry and an advisory shield.

Full-size SVG. Agent1's All-In-One policy has 20 categories with 20 questions each and ten event buckets. Unknown and inapplicable answers do not count as passes. The local .env review inventories variable names without returning values. Adding a GitHub Actions secret requires explicit target selection and hidden terminal entry; Actions secrets are write-only and cannot be read back to a local .env through a symlink. Local execution needs a separately chosen local secret source. Findings are advisory by default and do not prove a device is malware-free.

Access Matrix and Agenda

A canonical JSON decision record flows to a human-reviewed spreadsheet and a concise session context card.

Full-size SVG. The selected scope's GLOBAL/ACCESS-MATRIX.json is the machine-readable decision record. The GLOBAL/ACCESS-MATRIX.xlsx workbook is a human review surface. Its role rows do not themselves grant access. Agent Alpha's Agenda starts empty; the session-settings view loads selected status and rule-file paths and hashes, without secret values. Actual access follows the host, repository, and provider controls in force at the time of an action.

Concise agent role cards

Role Image and short alternative Detailed boundary
Agent 0 (agent.0) Coordinator and acceptance: a human objective and evidence return meet a central task ledger. Owns task planning, integration, and acceptance. External actions still require actual host authority.
Agent X (agent.x) Events and time: timestamp marker, calendar and clock, and due-item card. Records observed events and due items; it does not run a scheduler or deliver notifications.
Agent Y (agent.y) Bounded CLI delegate: assigned folder, terminal branch, and evidence package. Executes a selected installed provider task within host permissions; no merge or task-acceptance authority.
Agent Z (agent.z) Evidence reviewer: source files, magnified checklist, and advisory scorecard. Applies its versioned 100-question policy; its scorecard is not task acceptance or certification.
Agent Alpha (agent.a) Reference selector: request card, ten reference tabs, and concise result. Routes selected AUTOBOT references. Agenda starts empty and runs no scheduler or extra model workers.
Agent Beta (agent.b) Named workspace recorder: task record, separate workspace folders, and disconnected cloud. Records BOT/DOT identity and scope without a provider session, login, or grant.
Agent1 (agent.1) Security and access reviewer: names-only inventory, assessment shield, and guarded terminal. Advises on 400 questions, migration, and secrets; no device-clean guarantee.

Detailed agent infographics

Seven additional 7680 × 4320 SVGs give each role a readable record location, example prompt, CLI family, process flow, and authority boundary. Their labels and icons are selectable vector content over generated art. Slash-separated CLI names denote alternatives, not one pasted shell command. The full text companion provides the exact location, project-local help command, example prompt, sequence, and limits for each role. On macOS/Linux the launcher is bash .SYSTEMX/SYSTEMX.sh; on Windows PowerShell it is pwsh -File .SYSTEMX/SYSTEMX.ps1.

Agent Detailed stock SVG 8K release JPEG
Agent 0 Folder, prompt, CLI, flow and boundary 8K JPEG
Agent X Folder, prompt, CLI, flow and boundary 8K JPEG
Agent Y Folder, prompt, CLI, flow and boundary 8K JPEG
Agent Z Folder, prompt, CLI, flow and boundary 8K JPEG
Agent Alpha Folder, prompt, CLI, flow and boundary 8K JPEG
Agent Beta Folder, prompt, CLI, flow and boundary 8K JPEG
Agent1 Folder, prompt, CLI, flow and boundary 8K JPEG

The source artwork and derived 8K exports have separate identities. The checksum asset records the published 8K bytes and digests; authenticated downloads matched. No role image starts a worker, authenticates an account, or grants permission.

Agent 0 and the user retain canonical task acceptance. A native harness subagent or a named BOT instance can be mapped to a TASK ID, but neither becomes one of these seven role registrations merely by appearing in a diagram or workbook. The Access Matrix guide and Agent Beta, Agent1, and Agenda specification provide the operational details.

DOT SYSTEMX · User manual

1. Start here

2. Work and continuity

3. Setup and maintenance

4. Advanced reference

5. Project information

Repository · Use this template · MIT license

Clone this wiki locally