Stock agents guide

Stock agents are templates, not finished employees. Use them to accelerate setup, then adapt the Charter, tools, thresholds, and Steward chain to the real seat.

Where to create agents

Add agents from the unified agents surface. In the app, open it from the Agents item in the left sidebar's Define section, or from an Add agent action on the Responsibility Graph. The first screen asks for a creation mode — use a template, design a custom agent, or connect an existing agent — then walks the same conservative path for every mode: seat placement, Steward, draft Charter, manifest sign-off, sandbox dry run, and human go-live. Each step is command-backed, so the same workflow runs from the app, the CLI, and MCP.

Template review

Before dry run, review the template's responsibilities, tool assumptions, escalation rules, and measurables. Remove any authority the company is not ready to grant.

Template catalog

Seven stock templates ship today. Every one is draft-first, escalates the actions in its safety boundaries, and keeps spend, external sends, and credential scope human-owned until a Steward sets a threshold. Read the live list with agent_template.list / rost_list_agent_templates or the rost://agents/templates resource.

AP Clerk — `stock/ap-clerk`

• Function: Finance. Tier: 1 · Launch.
• Responsibilities: Ingest and match invoices; Route payable exceptions; Prepare payment drafts.
• Default tools: Accounting bills read/write drafts; AP inbox read/draft.
• Safety boundary: vendor bank changes, payment release, disputed or duplicate invoices, and unclear approval chains escalate before action.

AR & Collections Clerk — `stock/ar-collections-clerk`

• Function: Finance. Tier: 1 · Launch.
• Responsibilities: Monitor aging; Draft cadence reminders; Flag disputes and overdue accounts.
• Default tools: Accounting invoices read; Email send-as collections drafts.
• Safety boundary: customer disputes, legal threats, credits, account holds, and strategic accounts escalate; sends and term changes need approval.

Inbox & Calendar Assistant — `stock/inbox-calendar-assistant`

• Function: Admin/EA. Tier: 1 · Launch.
• Responsibilities: Triage shared or executive inbox; Draft replies; Propose meeting times and maintain follow-ups.
• Default tools: Gmail read/draft; Calendar read/write proposals.
• Safety boundary: sensitive content, VIP conflicts, external sends, deletes, and unclear authority escalate.

Customer Support Triager — `stock/customer-support-triager`

• Function: Support. Tier: 1 · Launch.
• Responsibilities: Classify and route tickets; Draft first responses from KB; Escalate angry or at-risk customers.
• Default tools: Helpdesk/support inbox read/draft.
• Safety boundary: angry customers, churn risk, outages, security/privacy, refunds, and policy exceptions escalate; sends and refunds need approval.

SDR Researcher — `stock/sdr-researcher`

• Function: Sales. Tier: 1 · Launch.
• Responsibilities: Enrich inbound leads; Build account briefs; Maintain CRM hygiene on new records.
• Default tools: CRM contacts read/write drafts; Web research.
• Safety boundary: sensitive personal data, regulated accounts, paid data, outreach sends, and unclear consent escalate.

Content Drafter — `stock/content-drafter`

• Function: Marketing. Tier: 1 · Launch.
• Responsibilities: Draft posts, newsletters, and pages from briefs; Maintain content calendar; Escalate claims and approvals.
• Default tools: Docs write to drafts folder.
• Safety boundary: unsupported claims, regulated topics, testimonials, pricing, legal, and any publication action escalate.

QA Reviewer — `stock/qa-reviewer`

• Function: Ops/Eng. Tier: 2 · Fast-follow.
• Responsibilities: Review release evidence; Draft risk issues; Escalate blockers before release.
• Default tools: Release artifacts read; Issue tracker draft.
• Safety boundary: release approval, production changes, customer-impacting risk, security/privacy, and conflicting evidence escalate.

Create from a template (CLI or MCP)

1. List templates: agent_template.list / rost_list_agent_templates with {} (scope: tenant). 2. Create a draft: agent.create_from_template / rost_create_agent_from_template with seat_id and template_slug. This is draft-only; occupancy still needs a human Steward. 3. Or run the guided path: agent_setup.start (mode template) returns a setup_id; refine with agent_setup.get and agent_setup.update. 4. Sandbox: agent.run_dry_run / rost_run_agent_dry_run after the manifest is signed.

For fully custom seats, see the custom agents guide (agent.create_custom).

Connect an existing or local agent

The third mode registers an agent you already run — Claude Code, Codex, Cursor, or a local Runner — against a seat instead of creating a new runtime. The seat, Steward, and Charter rules are identical; only the runtime is external.

1. Choose Connect an existing agent and place the seat with a human Steward (an agent or hybrid seat). 2. ROST mints a seat-scoped MCP token or a Runner work-order setup, both token-backed and redacted. Install it with rost mcp install --client <client> --scope seat --seat-id <seat-id>, or pair a Runner with runner.pairing.start (see the local runner guide). 3. The external agent verifies it can call rost_get_context and report status before setup is marked complete.

A seat-scoped token operates only as that one seat, bounded by the seat Charter and permission manifest; it can never create or approve its own setup, sign its own manifest, or go itself live. See the CLI and MCP installation guide for client registration and the local runner guide for Runner lanes.

When to stop for confirmation

agent.create_from_template, agent.create_custom, and agent.go_live are human_required; template listing and agent.run_dry_run are not gated. A template can draft and dry-run; only a human approves the Charter, permission manifest, Steward chain, and go-live. See the confirmations guide.

Go-live rule

Templates may draft. Humans approve. A stock agent should not go live until a human approves the Charter, permission manifest, and Steward chain.