Repository navigation
Agent Beta Agent1 and Agenda
Open the current Agent Beta guide · Open the current Agent1 guide · Open the current Agent Alpha guide
Verify the selected release and project-local command help before operating a project.
.SYSTEMX keeps one project task authority while offering three optional
extensions:
| Extension | What it adds | What it never implies |
|---|---|---|
Agent Beta (agent.b) |
Named BOT/DOT workspaces and canonical task references in AGENTB/; an online-reference may retain nonsecret provider/identity aliases and an access-decision reference. All current records remain offline. |
A connected provider, synchronized remote workspace, model process, or credential permission. |
Agent1 (agent.1, security-access) |
A guided local security/access assessment, safe credential-name inventory and migration plan, plus an explicitly started read-only stdio session. | A secret vault, GitHub admin role, automatic upload or rotation, permanent daemon, or task acceptance. |
| Agent Alpha Agenda | Ten fixed setting definitions and four access-decision states, with empty selected project options and decisions. | A scheduler or notification worker. Agent X owns observed events and due items. |
Agent 0 remains the coordinator. WORK/TASKS.json remains the only task-status
and acceptance record. Agent Y's CLI jobs, Agent Z's reviews, Agent X's event
history, and Agent Alpha's reference requests keep their existing boundaries.
Activating a role record does not start an AI provider or grant IAM rights.
Project-local role, security, Agenda, and service commands below are
subcommands of the selected project's launcher:
bash .SYSTEMX/SYSTEMX.sh COMMAND on macOS/Linux or
pwsh -File .SYSTEMX/SYSTEMX.ps1 COMMAND on Windows PowerShell. The installed
systemx package command manages installations; it does not accept a direct
systemx agent-1 or systemx agent-beta form.
Use the exact .SYSTEMX name. Choose the containing root or an explicitly
registered Projects/NAME/.SYSTEMXP child; do not infer the scope from the
last chat or current shell directory. The managed first-run command previews
the installation and empty configuration before --apply. It installs no OS
service or scheduler. See First-Time Setup
and the operator setup page
in the tagged beta source.
An Agent1 first-run review asks for the project owner, host and OS, actual owner/guest roles, GitHub account and repository, permitted Actions use, credential names and destinations, suspected exposure, and backup/cleanup plan. Unknown answers remain unknown. Beta's online workspaces and Alpha's Agenda have stock definitions, but selected options and decisions start empty and require separate review.
GitHub Actions secrets serve workflows in repositories where Actions is
enabled; they do not configure a local shell, keychain, .env file, or Agent1
session. GitHub Actions is disabled in the dotSYSTEMX template repository and
no beta workflow is included. Local credentials belong in an
approved OS credential store or other selected local secret manager. An ignored
.env file is still plaintext. Actions variables are suitable only for
nonsensitive values.
Choose repository, named environment, or organization scope deliberately.
Same-name precedence is environment, then repository, then organization.
Environment approval/branch controls depend on actual settings and plan.
gh secret list reports names and scopes, not values. GitHub's
Actions secrets guide,
secrets reference,
and CLI secret manual provide the
current provider behavior.
For .env migration, inventory key names privately; classify each as secret,
public setting, or obsolete; optionally preview a names-only .env.example
with agent-1 env-sanitize; choose a destination for each; confirm the active
account and exact repository/environment before any import; verify an actual
consumer without echoing the value; rotate at the source if exposure is
suspected; and only then retire old plaintext under the owner's retention
policy. env-migrate defaults to a newly issued credential intent. Reuse of
an existing test credential requires --credential-class existing-test and
--ack-test-only; this assertion does not verify rotation. No first-run action
bulk uploads or removes a .env file. See the
security spec.
For an explicit repository or named-environment secret set, Agent1 verifies
the active GitHub account's repository admin permission before hidden input.
An owner/operator/guest setup label is descriptive; verified guest admins may
proceed. It reads back secret name/scope metadata only, returning unknown
when readback cannot verify the name. It does not verify a value or workflow
consumer. Organization secret writes are outside this beta CLI.
A .SYSTEMX Access Matrix ships 20 standard role rows with templateGrant: none and requiresReview: true: seven agent roles, ten AUTOBOT references,
and instance patterns for numeric workers, native subagents, and Agent Beta
BOT/DOT workspaces. Its reviewed decisions array starts empty. The
Agenda settings catalog
defines ten context_selection choices, all initially unset; the
Access Decision Rules catalog
defines no-row, gap, denied, and reviewed-intent without making a
project decision. Inspect both definition files with
bash .SYSTEMX/SYSTEMX.sh agenda catalog; inspect selected choices and
revisions with bash .SYSTEMX/SYSTEMX.sh agenda status. On Windows PowerShell,
use pwsh -File .SYSTEMX/SYSTEMX.ps1 agenda catalog. An include needs a
same-scope reviewed-intent decision and a current Rules digest; exclude is
an explicit reviewed choice. Neither state grants external access.
Use agenda status to obtain accessRevision and agenda decision-add with that
revision to preview or apply one selected-scope intent, denial, or gap. The
decision --action must be nonempty; an empty value fails validation without
saving a row. The
checked-in XLSX has five stock sheets, including an Agent 0 roster; live CSV
fields from agenda table --format csv reflect scoped JSON records and fixed catalogs. An explicit
agenda workbook-export writes a current, formula-free six-sheet workbook,
including an Agent 0 roster tab, under ignored local/, guarded by the
source revision; agenda workbook-check detects drift. None is an OS or
GitHub grant. An adopted older matrix can preview agenda roles-upgrade to
add the stock role rows without replacing decisions, then apply it with the
observed revision.
For a named BOT/DOT, create its Agent Beta record first, then a same-actor
reviewed-intent row for agentb:BOT_NAME. Preview agent-beta access-link BOT_NAME --access-decision-id ACC-001; repeat with its combined BOT/matrix
revisionBefore as --expected-revision REVISION --apply. The reference
remains offline metadata and grants no external permission.
A guest with repository Write access can alter workflows and may arrange use or exfiltration of Actions secrets. If a guest must not reach those secrets, use a less privileged role or a separate repository and verify actual branch and environment controls. A nested role label cannot provide that separation. See GitHub's secure-use guidance and repository roles.
session-settings returns a single selected-scope JSON packet with Agent
Beta, Agent1, and Agenda status plus fixed AUTOBOT and registered Beta Rules
paths and SHA-256 hashes. It does not return rule contents, secret values,
provider accounts, or live permission state. Select a child with
projects session-settings --project NAME. Use --compact to return only a
rule count, absent IDs, and a revision digest when file-level paths are not
needed. On the blank root template this reduced output from 3,248 to 1,043
bytes; that is a byte count, not a billed-token benchmark.
The first Agent1 service is a short-lived stdio process. service status
reports local capability; service start --stdio starts the protocol for a
client that owns both pipes. It accepts fixed ping, capabilities, assess,
roster, settings, onboard, and stop methods. roster is Agent 0's
read-only local view; settings is the compact Rules summary/revision packet; onboard
returns default first-time navigation without a caller-selected provider or
dotenv path. It has no listening port, background installation, or
secret-field input. An assess response is advisory local evidence, not proof
of remote IAM, a working secret, or a running deployment. The
service protocol
gives exact framing and limits at the selected beta release.
The Agent 0 toolchain I/O guide explains generic
CLI/JSON-lines use and the separate host-integration boundary; this transport
is neither MCP nor ACP.
Host coverage: On macOS/Linux, dotenv and answer-file operations use
anchored no-follow handles and a one-megabyte post-open bound. Reconfiguring an
existing Agent1 setup requires the complete eight-answer preview, then
--apply --replace-existing --expect-sha256 with the observed digest. Windows
fails closed for dotenv/answer-file operations and setup reconfiguration;
metadata first-run setup/status/questions and the advisory filename scan
remain available. These are source behavior boundaries, not a verified
secret-consumer or device-security result.
Warnings about missing, stale, or unsafe facts are on by default. Alpha's
Agenda view does not dispatch a warning or job on a timer. Agent X due can
show due items on request; an external scheduler, if separately authorized,
owns actual recurrence, timezone rules, retries, and notifications.
Stop active writers and the stdio client, inspect external resources, then
preview systemx uninstall --target PROJECT --backup NEW_EXTERNAL_PATH.
--apply moves the entire .SYSTEMX folder to that new same-filesystem backup
and records an inventory/receipt. systemx audit --target PROJECT checks only
the selected project footprint. systemx restore previews and verifies a
return into an empty slot. The folder operation does not remove a Python
environment, GitHub secret, provider job, PR, cloud copy, or separately
installed scheduler. The operator uninstall page
and core uninstall guide
describe those scopes at the selected beta release.
.SYSTEMX · DOT SYSTEMX · Maintained by Wayne Tech Lab LLC · Repository · User manual · Prompt cookbook · MIT license
Beta 1.11.0-beta.3 prerelease — validate the selected release in your project. Release policy. Project records, configured commands, and verified evidence determine the state of each adopted project.
DOT SYSTEMX · User manual
1. Start here
- The Guided Folder: 4K walkthrough and visual chapter index
- Start with a Folder: 4K animated guide
- Introduction video and archived poster
- Advanced walkthrough: inside the coordination layer
- Choose a workspace
- Agent 0 roster: roles, bots, and CLI paths
- Agent 0 and toolchain direct I/O
- First-time setup
- Provider and harness onboarding
- Daily workflow
- One-shot project guide
- Shared workspaces, clouds, and bots
- Historical use-case workflows
- Four current system maps
- Current maps, seven agent guides, and atlas
- Agent guide text and CLI flows
- Visual guide and historical diagrams
- New chat setup prompt
- Prompt cookbook
2. Work and continuity
- SYSTEMX PROJECTS
- AUTOBOT and Agent Alpha: Tier 2
- Local DATA storage
- Planning and memory
- ROOT planning, rules, and task delta
- System MAP
- Access Matrix: 20 standard entries
- Agent Beta, Agent1, and Agenda
- Task workflow
- Agent X: events and time
- Agent Y: CLI and PR channels
- Agent Z: fixed review
- 100 review questions
- Agent 0 and subagents
- Subagent processing
- Advanced use cases
- Evidence and acceptance
- Anti-drift and long-running work
- Tokens, cost, and time
3. Setup and maintenance
- Project / VS Code
- OS directory
- Google Drive
- LLM chat
- Exact casing and alias
- Installation and updates
- Upgrading
- Uninstall and cleanup
4. Advanced reference
- Standard format
- Technical Guide
- Stack Guide
- Library integration
- CLI and Tools
- Command reference
- Configuration and quality
- Operations and troubleshooting
- Security and support
- Code review and readiness
5. Project information