Repository navigation
Four Part System Infographic
The current .SYSTEMX field guide is four separate 3840 × 2160 pages:
one full-system overview followed by three detailed views. Read page 1 for the
connections, then pages 2–4 for execution, reference routing, and storage/scope
boundaries. Click any image to open its full-size export.
The refreshed edition depicts the 1.11.0-alpha.2 feature contract, including ROOT planning/rules and observed task-list reconciliation, for the 1.11.0-alpha.3 media/manual release. A diagram is a navigation aid, not proof of a running worker, a permission grant, or a deployed application. First-time setup explains installation before use.
| Page | Read it to answer |
|---|---|
| 1. Full-system overview | How do ROOT, task-list observations, Agent 0/X/Y/Z, AGENTA, all ten AUTOBOTS, DATA, and .SYSTEMXP connect? |
| 2. Agent coordination and CLI tooling | Who assigns/executes/reviews a task, compares external TODO observations, and reconciles through authorized tools plus readback? |
| 3. AGENTA and the ten AUTOBOTS | How do selected specialist references and ROOT internal rules support the task without creating a second backlog? |
| 4. DATA and named project scopes | Where do private payloads go, and how do each project’s ROOT, canonical tasks, and external mappings stay separate? |
Current artwork: Complete system with ROOT.
Open / download the full 4K JPEG
Text equivalent: start with one authorized objective in one explicit scope.
Global context and the Master Plan define accepted purpose and outcomes;
WORK/TASKS.json owns task status and evidence. Optional ROOT/SX-RULES.md
defines internal coordination conventions, ROOT/SX-MILESTONE.md explains
planning/tool cards, ROOT/PLAN.json holds request/structure references, and
ROOT/MIRRORS.json retains external bindings/observations/cards. These records
link the selected task ledger to IDE/harness and optional cloud/bot task items
without duplicating canonical status. Agent 0 routes work, integrates
results, and retains acceptance with the user. The harness supplies real tools,
processes, permissions, and native workers. Agent X records actual events/time;
Agent Y carries a bounded task through one selected installed external CLI;
Agent Z applies the same versioned 100-question policy. Agent Alpha (AGENTA,
agent.a) returns selected reference detail from ten named AUTOBOTS. DATA keeps
local payloads behind default Git-ignore rules, with exact scaffold exceptions.
Named .SYSTEMXP children share outer tools while retaining separate records.
Useful detail remains in its canonical home. CURRENT.md and the six work-state
views are generated from task/focus records; Memory and Archive do not create
competing status lists. Select only the context needed for the current task.
The IDE and an optional bot keep their actual external task lists; ROOT retains
observations and a read-only delta. Agent 0 uses authorized tools to reconcile
the affected surface, reads back the result, and records the observation.
Provider completion does not accept a canonical task.
ROOT planning and task delta ·
Planning and memory · Task workflow ·
Standard format · System MAP.
Current artwork: Three TODO surfaces, one task authority.
Open / download the full 4K JPEG
Text equivalent: three surfaces meet through scoped task-ID bindings:
canonical WORK/TASKS.json, an actual IDE/harness plan snapshot, and an optional
bot/run snapshot. ROOT retains raw/normalized states, observation/recording time,
source revision, evidence, and canonical fingerprints. Revisions remain
caller-supplied claims. Agent 0 selects affected tasks, reads actual external
state, retains a snapshot, runs read-only delta, reconciles through authorized
tools, reads back the result, and checkpoints.
Agent 0 also connects the originating request to a milestone, related waves/phases, and bounded canonical tasks. Optional hierarchy levels are expanded only where useful; fixed counts are not required. Each assignment identifies its scope, task ID, source revision, permitted files, dependency boundary, acceptance evidence, budget, and return format. Independent tasks may use authorized app/IDE workers or an explicitly selected Agent Y CLI lane. The two execution lanes need separate write scopes; a CLI role is not another invisible app subagent.
| Role or tool | Responsibility and boundary |
|---|---|
| Agent 0 | Maintain plan/tasks, select bounded work, examine task delta, reconcile through authorized real tools and readback, and retain acceptance. |
| App / IDE workers | Execute scoped work with tools and permissions actually supplied by the harness; return changes and original evidence. |
| Agent X | Record events, actors, occurrence times, attempts, revisions, evidence, and due items. External tools execute schedules. |
| Agent Y | Use one task-selected, already installed CLI provider and isolated Git scope; return its commit/worktree and optional opted-in draft PR. |
| Agent Z | Review the fixed policy of 10 categories × 10 questions; preserve unknowns and evidence. A score does not accept a task. |
| Agent Alpha / AGENTA | Select relevant AUTOBOT reference material and retain bounded request/result pointers. It does not execute the task. |
| ROOT tools | Save scoped request/structure/card references and external observations; compare retained snapshots without contacting or updating a harness/bot. |
| Management CLI / local launcher | Select defaults, installation or root/child records, and explicit commands; it does not install a model provider. |
| Configured project tools | Run actual check/dev/build/deploy commands when configured and authorized, retaining real exit status and evidence. |
Agent Y's optional draft PR is a handoff. Accepted-ready state, cancellation of an unmerged PR, and cleanup of a verified merged-and-accepted branch are separate lifecycle decisions. No review score grants merge authority. Agent 0 cites or runs independent project checks before acceptance and retains a precise remaining delta when criteria are unmet. Save reviewed facts and a checkpoint; do not repeat unchanged reviews or open a new work loop merely because an event was recorded.
Three task surfaces need mapped agreement about the same work, rather than equal row counts: canonical SYSTEMX tasks, actual harness items, and optional bot/provider items. One item may map several tasks; one task may have several attempts. The delta compares the aggregate state and fingerprints of all bound tasks, including members outside the selected output. It reports unmapped, unavailable, unverified, stale, or divergent observations. Default assessment is harness-only; explicitly include a bot layer when one is actually used.
Agent 0 and subagents · Agent X · Agent Y · Agent Z · CLI and Tools · Evidence and acceptance.
Current artwork: ROOT planning and on-demand specialists.
Open / download the full 4K JPEG
Text equivalent: ROOT defines internal rules and the planning/tool-card
vocabulary below actual user and host instructions. PLAN.json links RTask
origin and bounded STask child/follow-up/delegation provenance to structural
nodes. The flexible outline is milestone → wave → phase → optional group →
task → optional subtask → single action; task status remains canonical.
MIRRORS.json retains bounded tool cards and observations. A saved card
describes a call; actual authorized harness tools perform it. ROOT packets allow
at most ten selected tasks and 64 KiB; other responses allow 128 KiB.
Agent 0, X, Y, Z, or an explicitly configured Tools step submits a scoped purpose
and input revision to Agent Alpha. Alpha selects only
the necessary fixed slots; the coordinator is outside the ten-specialist
limit. Exact file names, including uppercase .MD, are part of the format.
| Specialist | Exact file | Selected project detail |
|---|---|---|
| Design | Design.MD |
Layout, user journeys, components, responsive behavior, accessibility, design rationale |
| Soul | Soul.MD |
Product purpose, audience, values, voice and experience; no claim of consciousness |
| Memory | Memory.MD |
Bounded working summaries and pointers to accepted durable project facts |
| Rules | Rules.MD |
Project conventions, cited policy, constraints, exceptions and precedence boundaries |
| Research | Research.MD |
Questions, original sources, observation dates, corroboration and uncertainty |
| Brand | Brand.MD |
Legal/trading/product names, identity, positioning and approved brand-kit references |
| Media | Media.MD |
Logo/ICO/favicon inventory, dimensions, variants, provenance, rights, licenses and exports |
| Tokens | Tokens.MD |
Semantic design tokens, CSS variables, typography, themes, spacing, states and contrast |
| Data | Data.MD |
Classification, ownership, definitions, schemas, retention and safe payload pointers |
| Archive | Archive.MD |
Superseded-reference index, dates, predecessor identity and recovery context |
AGENTA.MD is the routing charter, CATALOG.json identifies the fixed panel,
and REQUESTS.json retains selected requests/results. Initialize agent.a
explicitly; X/Y/Z roles-init is separate. The public template starts with only
Agent 0 registered and blank reusable specialist records.
A request fingerprint includes its selected local reference inputs and declared
revision. Identical requests reuse the saved identity and any completed result,
including original evidence time. Changed inputs justify a new bounded request;
identical completion is reusable and a conflicting second completion is refused.
A deliberate refresh needs an honest new --input-revision, not a polling loop.
The tools cannot independently establish that a caller's external revision is true.
autobot context reads only the charter and selected specialist documents, with
a 64 KiB packet limit and 256 KiB per specialist file. It does not read
DATA payloads or fetch URLs. An authorized human/LLM performs any actual research,
file edits, or model calls through the available harness. Rules and imported
research remain lower-authority project data; caller labels are attribution,
not authentication. Request completion never changes task acceptance.
ROOT/SX-RULES.md governs internal coordination conventions; the selected
Rules.MD specialist holds project/policy reference detail. Link the applicable
source and scope rather than copying every rule or task status into both. ROOT
planning/tool cards may request relevant context, and Agent Alpha reuses matching
results with their original evidence time. Neither layer automatically calls
all specialists, runs a review, or creates another TODO authority.
AUTOBOT and Agent Alpha contains activation, request, completion, reuse, and child-scope command examples.
Current artwork: Scoped ROOT, DATA and the toolchain.
Open / download the full 4K JPEG
Text equivalent: one outer installation provides selected standards, tooling,
version policy, and a registry. Explicitly select the root or one named child:
.SYSTEMX/Projects/Project-A/.SYSTEMXP. Project B has its own named child path.
Each owns separate context, plan, task ledger, focus, memory, and optional role,
AUTOBOT, ROOT, and DATA records. Each initialized ROOT layer has its own
SX-RULES.md, SX-MILESTONE.md, PLAN.json, and MIRRORS.json. Task IDs and
external bindings are local to that selected scope. Children do not inherit parent/sibling private facts,
request results, brand, or payloads. Do not create nested .SYSTEMX, a nameless
Projects/.SYSTEMXP, a second manager, or recursive child Projects trees.
| DATA folder | Intended local payload use |
|---|---|
TEMP/ |
Temporary work and scratch inputs |
LARGE/ |
Large binaries or datasets |
IMPORTS/ |
Authorized incoming originals |
RESEARCH/ |
Local source material and working research data |
ASSETS/ |
Private/draft brand, design, fonts, images and media |
EXPORTS/ |
Generated local outputs |
CACHE/ |
Rebuildable local cache payloads |
ARCHIVE/ |
Retained superseded payloads and recovery material |
Only exact README/policy scaffolds are intended to be tracked; payloads are
ignored by Git by default. data audit inspects sampled ignore rules and the
tracked index without reading payload contents. It is not exhaustive verification
of every actual path, history, another clone, or a custom unignore exception.
Already tracked/force-added files remain publishable. Check the actual path and
tracked state before exporting anything sensitive. Ignore rules do not govern
Drive/cloud/backup sync, encryption, access permissions, or credentials.
A named source URL describes a location; it does not grant access or transport bytes. Git/Drive/cloud connectors need real authorized access, one record writer, conflict handling, saved revisions and readback. Child setup and updates add missing scaffolds without replacing adopted records or deleting local folders. A cloud/chat URL provides a pointer; saved observations are not a live connector. Use real authorized transport and readback to carry records between hosts, then resolve the selected scope’s delta without merging unrelated project ledgers. TEMP/CACHE names authorize no cleanup. Whole-install uninstall/restore includes local DATA and the Projects tree in the reviewed external backup, so stop writers and inspect the preview before a lifecycle change.
SYSTEMX PROJECTS · Local DATA storage · Shared workspaces, clouds, and bots · Uninstall and cleanup.
The four selected native imagegen outputs are 1672 × 941 pixels. Their
shipped 3840 × 2160 JPEG exports use macOS sips, JPEG quality 85.
They are upscaled: the 4K canvas adds pixels without creating native detail.
Only the selected 4K exports ship; original native dimensions and exact-byte
fingerprints remain in metadata. Final technical/visual review checked the
ROOT paths, five role boundaries, ten specialist slots, independent child
records, observed task mappings, read-only delta, and authorized readback.
Generation and targeted technical edits are retained in the prompt record.
The refreshed JPEG outputs are linked to the fixed v1.11.0-alpha.3 tag so
this edition can be reproduced. Exact native/export dimensions, method, byte
sizes, SHA-256 checksums, and prompts are retained in the
asset metadata
and media reference library.
The portable text companion
and generation prompts
travel with the release. Use the command manuals for exact syntax and limits.
| Page / final artwork title | Bytes | SHA-256 of exact 4K JPEG |
|---|---|---|
| 1 — Complete system with ROOT | 1,900,067 | 932b114f65f277ebce216f47795ec03c1e8001c749ba63f4f3840168a4b6d7e6 |
| 2 — Three TODO surfaces, one task authority | 1,893,890 | 9950a100561b4c30d2aad232865af09fa9a3a51d142e004291e76c83a37584e8 |
| 3 — ROOT planning and on-demand specialists | 1,809,511 | 4f44e03bd7103e0464f87af063b97a312e985f44d0d86ca81368e9a9345c049f |
| 4 — Scoped ROOT, DATA and the toolchain | 1,919,070 | 86118223ad7cdd34ab74cd0b541dc73d0f274ad4915ef9db6df037989175e966 |
The earlier 1.10 field-guide edition remains available at its immutable tag. New release assets and URLs identify the refreshed ROOT artwork; preserved local project files may still show an older edition after an additive update. Earlier environment cards retain their historical baseline. The current edition 2.0 white paper examines alpha.4 and supplies four separate publication figures; this four-image raster series retains 1.11.0-alpha.3 media edition and its exact bytes.
Read ROOT planning, rules, and task delta for the
editable hierarchy and task-surface diagrams, first-use setup, tool cards,
selected packets, freshness checks, and role boundaries. root delta compares
retained local observations; it never calls an IDE/provider API, executes a
worker, accepts work, or writes an external TODO list. An actual reconciliation
uses available authorized tools, verifies their result, and retains readback.
Unknown access and stale evidence remain explicit. A current diagram cannot
prove a live external connection or downstream production acceptance.