Repository navigation
Provider and Harness Onboarding
Select a project scope and a real host tool before assigning work. A .SYSTEMX
role or named BOT workspace is a coordination record. The provider's own login,
account permissions, model and tool availability, data-handling terms, and
runtime state must be checked in that provider. Agent 0 keeps the canonical
WORK/TASKS.json task and accepts evidence after the run.
- Choose the containing
.SYSTEMXor a registered.SYSTEMXPchild. Record the objective and first accepted task. Read the selected release'sSTART-HERE.mdand first-time setup. - Review Agent1's security and access setup
for owner/guest role, exact repository and environment, local secret source,
.envnames-only inventory, and rotation intent. Never paste secret values into a shared task, workbook, bot record, or assistant log. - Inspect the Agent 0 roster for local role, BOT, AUTOBOT, recorded CLI-channel, and access-decision references. A roster row does not prove an external account or permission.
- Choose the actual lane below. Check its installed tool or connected app, active account, project access, data boundary, and supported actions. Record the result of a real test in the selected project's evidence, with the revision and provider identity needed to review it.
- Delegate one bounded task, return the provider's artifact and actual readback, and have Agent 0 or the user decide acceptance. A provider's completed status alone does not close the canonical task.
From the containing project directory, the selected-scope onboard views
combine Agent1 question counts, Agent 0 roster counts, and a provider navigation
catalog. They return no credential values and do not authenticate, install, or
run a provider:
bash .SYSTEMX/SYSTEMX.sh onboard status
bash .SYSTEMX/SYSTEMX.sh onboard plan \
--account-context personal --surface cli --provider codex-cli \
--credential-consumer local-cli --credential-class new
bash .SYSTEMX/SYSTEMX.sh onboard plan \
--account-context enterprise --surface sdk --provider codex-sdk \
--credential-consumer server-application --credential-class newChoose enterprise and the actual consumer when those are the facts. The
plan response lists unresolved first-time questions and the selected
provider's next checks. installedNameOnly means an executable name was
found on this host; it does not establish login, model access, permission, or a
safe task run. A null name-check result means the catalog has no safe
executable probe, as with SDK/IDE/cloud surfaces or Cursor's generic agent
command. The default output is readable text; use --format json for a
machine-readable packet that includes installedNameOnly. With an explicitly
selected existing dotenv file, --env-file
uses Agent1's bounded names-only inventory on supported hosts. Windows fails
closed for dotenv and answer-file operations in this beta. This view neither
imports nor converts a .env value automatically. Do not place values in
command arguments or project records. GitHub Actions secrets are for
workflows in the selected repository; they cannot supply a local CLI, IDE, or
.env through a symbolic link. Actions is disabled for dotSYSTEMX itself;
check the selected project's Actions setting and actual secret consumer
separately.
For a host that discovers project-root AGENTS.md, onboard bridge previews a
short pointer to the selected .SYSTEMX/AGENTS.md. It creates the root pointer
only after an explicit --apply --expected-sha256 SHA_FROM_PREVIEW using the
preview's contentSha256. It preserves an existing root instruction file and
does not configure Codex, Grok, Muse, or another provider. The preview works on
Windows, but applying the bridge requires anchored directory handles and fails
closed there in this beta. The bridge is for the outer .SYSTEMX project root,
not a .SYSTEMXP child. Menu options 34, 35,
and 36 open status, plan, and bridge preview respectively. The selected
provider's own host discovery and trust rules still govern whether it reads
that pointer.
Consumer selected in onboard plan
|
First-time check |
|---|---|
| Local CLI or VS Code extension | Prefer the provider's own login or an approved local OS credential store; verify the host account and project access. A GitHub Actions secret cannot fill either surface. |
| GitHub Actions | Check the selected repository's Actions setting, workflow, environment, scope, and actual consumer. dotSYSTEMX itself has Actions disabled and no workflow. |
| Codespaces, GitHub agents, or Dependabot | Use that product's own secret scope and verify the intended consumer there; a similarly named Actions secret is not interchangeable. |
| None | Record that no credential destination was selected; do not create or copy a key. |
Use a newly issued credential by default when one is needed. Treat an existing
key as test-only while the owner reviews rotation and downstream consumers.
An existing .env can be inventoried by key names and previewed as a
names-only .env.example through Agent1; the user then selects the actual
store or provider destination for each key. The current guide does not bulk
import, delete, rotate, or verify credential values. See
Agent Beta, Agent1, and Agenda.
| Lane |
--provider ID |
.SYSTEMX record |
Execution and account boundary |
|---|---|---|---|
| Codex CLI | codex-cli |
A reviewed Agent Y task-scoped handoff | The separately installed CLI and its login, selected model, repository permissions, and task run remain external. |
| Codex IDE extension | codex-ide |
Agent 0 task and optional harness mapping | The host IDE supplies sign-in, tools, and permissions; the onboarding view does not install or authenticate it. |
| Native subagent | No provider ID; choose its actual host | Agent 0 task and a decision-only native-subagent Access Matrix row when needed | The host app/IDE supplies worker creation, tools, account, and permissions. A roster pattern is not a registered worker. |
| OpenAI DOT | openai-dot |
Optional Agent Beta named dot workspace and canonical task reference |
A DOT is a separately created cloud agent. The local workspace is an offline nonsecret reference; local computer access is an additional opt-in in the host. |
| xAI Grok Build CLI | grok-build |
An optional manual external CLI lane | The grok executable has its own login and instruction discovery. Agent Y's configured adapter is not a reviewed Grok prompt-input contract, so this guide does not route Grok through it. |
| xAI Grok Bot | grok-bot |
Optional Agent Beta named bot workspace and task reference |
The persistent cloud Bot is separate from Grok Build CLI. The local record remains offline and does not connect the cloud Bot; xAI says Bots share a user's cloud computer, so do not assume isolation between them. |
| Meta Muse Code CLI | muse-code |
An Agent Y handoff after the adapter and actual CLI are reviewed | The muse executable, login, model, account rights, and task run remain external. |
| Meta personal Muse | meta-muse |
Optional Agent Beta named bot workspace and task reference |
The personal product is separate from Muse Code CLI; the local record remains offline. |
| VS Code or another IDE harness | Select its provider where relevant | Selected project records plus a host-assigned worker identity | The IDE and extension supply execution and their own permission controls. |
| GitHub Copilot CLI | copilot-cli |
A reviewed Agent Y task-scoped handoff |
agent-y providers detects the installed name; GitHub login and access still require a host check. |
| Reviewed configured CLI | configured-cli |
A task-scoped Agent Y configured adapter after an explicit absolute executable and argument/input contract are reviewed | Name discovery, login, tool permissions, and a safe noninteractive route are not supplied by this catalog ID. |
The onboarding catalog may list a surface before an Agent Y execution adapter
exists. These entries guide the user's next host check; they do not install,
sign in, run, or connect a provider. agent-y providers lists its supported
Agent Y CLI names separately from this broader onboarding catalog.
| Surface |
--provider ID |
Route and boundary |
|---|---|---|
| Codex SDK | codex-sdk |
Application-managed TypeScript/Python SDK integration through the developer's own application and credentials; no .SYSTEMX process or automatic SDK call. |
| Amazon Kiro CLI | kiro-cli |
Formerly Amazon Q Developer CLI. Manual/discovery kiro-cli terminal entry; no Agent Y task-execution adapter. The generic kiro name can route to the IDE. |
| Amazon Kiro IDE | kiro-ide |
Separate desktop IDE and its own sign-in, workspace, and permissions; no Agent Y process or VS Code extension claim. |
| Amazon Kiro web | kiro-web |
Browser cloud sessions require host sign-in and repository authorization; Agent Beta holds an offline reference only. |
| AWS CLI | aws-cli |
aws is a host account/infrastructure tool, separate from Kiro AI and outside Agent Y. Verify the selected profile and effective AWS identity. |
| Anthropic Claude Code | claude-code |
Manual/discovery claude CLI; no Agent Y execution adapter. |
| OpenCode CLI | opencode-cli |
Manual/discovery opencode CLI; check its actual V1/V2 binary, version, and provider account before use. No Agent Y adapter. |
| BLACKBOX CLI | blackbox-cli |
Manual/discovery blackbox CLI; no Agent Y execution adapter. |
| Gemini CLI | gemini-cli |
Manual/discovery gemini CLI; its own authentication source must be checked. No Agent Y adapter. |
| Cursor Agent | cursor-agent |
Manual/discovery CLI whose current command is agent; a generic agent on PATH does not prove Cursor identity. Separate from the Cursor IDE host; no Agent Y adapter. |
| Aider | aider |
Manual/discovery aider CLI; no Agent Y execution adapter. Review its model keys and automatic commit behavior; keep credentials out of command-line examples and logs. |
The tagged provider catalog is the release baseline for these provider IDs, surfaces, and executable-name checks. The official setup links above describe the separate host tools.
The host checks are different for each row. Kiro's own
authentication and
kiro-cli version/kiro-cli whoami are separate from an AWS CLI profile and
IAM Identity Center;
an AWS account identity is not a Kiro coding-agent session. BLACKBOX's
blackbox configure
can select internal provider modes; they stay inside one BLACKBOX CLI lane here.
An installed opencode name does not distinguish V1 from V2, and
V2 may start its own background server.
Gemini CLI authentication
may use Google sign-in, an API key, or Vertex; its .gemini/.env behavior does
not turn GitHub Actions secrets into local credentials. Keep Aider's API keys
out of example command arguments and transcripts. All of these are external
host behaviors, not effects of the read-only .SYSTEMX onboarding view.
The catalog is navigation and local discovery. A stored reference or matching executable name does not prove a live provider connection, account authority, or completed task.
OpenAI's Codex CLI, Codex IDE, and DOT getting started describe their distinct host setups. xAI documents Grok Build and Grok Bot separately. Meta documents Muse Code and its authentication choices; the personal Muse agent is a separate app/web/WhatsApp surface. The selected provider's own documentation remains the source for authentication and live permissions.
For a CLI lane, inspect its own login and instruction discovery before handing
off a bounded task. Codex documents codex login status
and repository-root AGENTS.md;
Grok Build CLI documents
grok login and grok inspect; Muse Code authentication
requires the selected workspace's trust and sign-in or approved API-key path.
These checks belong to the installed host. A local onboard or root roster
response cannot substitute for them. An IDE extension and a CLI inside VS
Code's integrated terminal are separate surfaces; choose vscode-extension
or vscode-terminal in the plan accordingly.
The project-local launcher is bash .SYSTEMX/SYSTEMX.sh on macOS/Linux and
pwsh -File .SYSTEMX/SYSTEMX.ps1 on Windows PowerShell. The installed
systemx command manages versioned installation and routes project commands
through run or projects; it is not an AI provider. The CLI and Tools
guide distinguishes it from Agent Y's selected Codex/Copilot/Muse Code/configured CLI,
GitHub CLI gh, IDE tools, and app-native workers.
agent-beta records a named BOT/DOT workspace without starting a provider.
root roster reports selected-scope pointers without probing an account.
agenda catalog shows stock reference settings and decision rules without
selecting project options or granting rights. An Access Matrix
reviewed-intent row records review intent, not an effective OS, repository,
or provider permission. See Shared workspaces, clouds, and bots
and Access Matrix for actual handoff and review boundaries.
For an IDE or harness that owns a local read-only process, the
Agent 0 toolchain I/O guide describes the fixed
JSON-lines route and the separate host-registration gate.
.SYSTEMX · DOT SYSTEMX · Maintained by Wayne Tech Lab LLC · Repository · User manual · Prompt cookbook · MIT license
Beta 1.11.0-beta.4 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