Agent reference map

Use this map when you are a CLI session, MCP client, or in-app agent and need to know which ROST guide to retrieve before advising a user.

Public URLs

• Human docs index: /docs
• Manifest: /api/reference/manifest
• Search: /api/reference/search?q=Charter&audience=mcp
• File: /api/reference/files/charter-design-playbook

MCP surfaces

• Search tool: rost_reference_search
• Get tool: rost_reference_get
• Reference resource template: rost://reference/{slug}
• Operation resources (read-only, scope-enforced): rost://tenant/status, rost://graph, rost://seat/{id}/context, rost://compass/current, rost://sync/latest, rost://operations/reference-map

Start with rost_reference_search when the user asks a broad question. Use rost_reference_get when a workflow or command metadata already names a guide slug. Read the operation resources to load current tenant/seat facts before mutating; tenant-wide resources need a tenant-admin token and a seat token may read only its own rost://seat/{id}/context.

CLI surface

• List guides: rost reference list
• Search guides: rost reference search "agent tools"
• Print a guide: rost reference get agent-staffing-playbook

Discover a command before calling it

Never guess a command's JSON shape. Before calling a command that changes state, read its exact contract:

• List every callable command: rost command list (CLI) or rost_list_commands (MCP)
• Read one command's exact input/output schema, help pointer, and a worked example: rost command schema <id> (CLI) or rost_describe_command with {"command_id":"<id>"} (MCP)
• List the tool catalog the agent builder reads (id, scope tiers, credential requirement, access policy — configuration only; the tools do not execute yet): rost tools list (CLI) or rost_list_tool_catalog (MCP)
• Show a Compass, Charter, or agent setup as a markdown card to review with your human: rost compass show --markdown, rost charter show --seat-id <id> --markdown, rost agent show --seat-id <id> --markdown

When a command fails, the error returns a machine-readable code, a message, and a help field naming the exact command to run next. Read the help field and run the command it points at — do not retry the same call blindly. A failed precondition (for example a manifest not yet signed, a dry run that has not passed, or the inference budget hard cap reached) returns COMMAND_PRECONDITION_FAILED with a help pointer, not an opaque internal error.

Inference budget

Agents that run on ROST-managed inference draw against a tenant inference budget. A new company starts with a hard cap of $0, so a real managed-inference run is blocked until a human raises the cap with settings.update (rost settings update --hard-cap-usd <amount>). Hitting the cap returns a typed COMMAND_PRECONDITION_FAILED whose details.reason is budget.hard_cap_exceeded, with a help pointer naming that exact next command — not a generic internal error. The sandbox dry run is free and is never blocked by the cap, so an agent can be chartered, dry-run, and taken live before any budget is set; only real runs are gated. See settings-guide.

Standard setup order

1. rost-implementation-method 2. compass-authoring-guide 3. responsibility-graph-playbook 4. charter-design-playbook 5. charter-authoring-deep-dive 6. agent-staffing-playbook 7. add-agents-guide 8. custom-agents-guide 9. agent-builder-guide 10. how-agents-work 11. tool-access-and-vault 12. available-tools-guide 13. mcp-and-cli-guide 14. cascade-guide 15. signal-guide 16. friction-guide 17. confirmations-guide 18. steward-queue-guide 19. sync-rhythm-playbook

Workflow to guide map

Read the listed guide before recommending or running each workflow. Every workflow ends at a human gate for any authority-changing step (see confirmations-guide).

• Connect a local agent (CLI login, MCP token, client registration): mcp-and-cli-guide, then agent-reference-map.
• Company onboarding (context, Compass draft, finish): rost-implementation-method, then settings-members-and-invites-guide for invites.
• Author a strong Compass directly (attach references, draft, answer gaps, approve): compass-authoring-guide — it shows the full JSON contract, what good vs weak looks like per field, and worked examples. Attaching reference docs (onboarding.attach_reference) is reference-only; you author the Compass, the server does not build it for you.
• Design the graph (create, rename, reparent, set type): responsibility-graph-playbook.
• Write Charters (draft, edit, approve, sign manifest): charter-design-playbook for the narrative, then charter-authoring-deep-dive for the full document contract field by field (decision_authority, escalation_rules, measurables, budget, permission_manifest, seat-type), then tool-access-and-vault for permissions. Submit the finished Charter directly with charter.update_draft or charter.set.
• Add an agent through the app (graph or sidebar, visual journey): add-agents-guide.
• Create an agent from a template: stock-agents-guide, then how-agents-work.
• Create a custom agent (operational answers, Charter Builder, configure tools, dry run, go live): custom-agents-guide, then agent-staffing-playbook and tool-access-and-vault.
• Build a controlled agent worker directly on the CLI/MCP path (seat, steward, job, boundaries, tools, credentials, model, schedule, dry-run, go-live): agent-builder-guide — it shows the full setup sequence, the structured model config, the access tiers, and how to discover tools (tool.catalog) and show your human a markdown readout (agent.show_markdown). The tool catalog is configuration the builder reads; the tools do not execute yet.
• Connect an existing or local agent to a seat: stock-agents-guide (connect section), then mcp-and-cli-guide and runner-guide.
• Run the operating loop as a seat (context, tasks, status, escalate): how-agents-work.
• Track goals: cascade-guide. Track measurables: signal-guide. File and resolve issues, tasks, escalations: friction-guide.
• Steward decisions (read queue, resolve, reject, approve go-live): steward-queue-guide, then confirmations-guide.
• Run the weekly Sync (compile, run, complete, assign): sync-rhythm-playbook.
• Operate local runners and work orders: runner-guide.
• Approve or reject a pending confirmation (human only): confirmations-guide.
• Manage members, settings, and notifications: settings-guide, settings-members-and-invites-guide, notifications-guide.
• Diagnose a failure on any surface: troubleshooting-guide, then security-model-guide.

Stage map

• Company setup: rost-implementation-method, compass-authoring-guide, settings-guide, settings-members-and-invites-guide
• Graph design: responsibility-graph-playbook
• Charter design: charter-design-playbook, charter-authoring-deep-dive
• Staffing: agent-staffing-playbook, add-agents-guide, custom-agents-guide, agent-builder-guide, how-agents-work, tool-access-and-vault, available-tools-guide, mcp-and-cli-guide, stock-agents-guide
• Operating rhythm: cascade-guide, signal-guide, friction-guide, confirmations-guide, steward-queue-guide, sync-rhythm-playbook, notifications-guide
• Local agents and runners: mcp-and-cli-guide, runner-guide, agent-reference-map
• Security and troubleshooting: security-model-guide, troubleshooting-guide

Agent behavior

Retrieve the narrowest relevant guide before making a setup recommendation. Prefer ROST terminology. Do not name internal repo paths in user-facing answers. Do not expose raw secrets, credentials, private tenant data, or internal planning documents.