Developers

Automation recipes

Safe patterns for workspace briefings, task intake, memory synchronization, employee setup, and custom MCP configuration.

These recipes describe bounded sequences. Inspect the live REST schema or MCP tool inputs before implementation.

Daily workspace briefing

Use get_workspace_overview, list_tasks, and get_runtime_health. Summarize blocked and in-review work, pending attention, employee ownership, and runtime freshness. Keep this key read-only.

Structured task intake

  1. Read list_agents and choose the intended owner.
  2. Normalize the request into a title, summary, priority, and labels.
  3. Call create_task once.
  4. Call read_task on the returned identifier to confirm assignment and status.
  5. Link the returned task identifier in the calling system.

Do not auto-retry an ambiguous failed write without checking whether the task already exists.

Review and reconcile existing work

  1. Use list_tasks with a single status such as REVIEW or BLOCKED.
  2. Call read_task before deciding; inspect comments, activity, approval summaries, and evidence-file metadata.
  3. Use comment_task to record the reviewer, evidence needed, or next action.
  4. Use update_task for content, assignment, priority, due date, or a status transition; use move_task when board ordering matters.
  5. Re-read the task and compare status totals after the bounded batch.

A task comment or status change is not approval for a linked send, post, publish, spend, deploy, delete, or other guarded action. Do not mass-close a review queue merely to reduce its count, and do not assume two UI totals represent the same filters.

Synchronize approved company context

  1. List current memory with list_memory.
  2. Match the source document to an existing Yodu memory record.
  3. Present a diff for human review.
  4. After approval, call create_memory for a replacement version, or update_memory when editing an existing draft. Direct active edits are validated and re-applied, but a replacement draft keeps the review boundary explicit.
  5. Call activate_memory with the reviewed draft identifier so the new version becomes active and reaches the runtime.
  6. Record the source version outside Yodu for the next comparison.

Avoid replacing a carefully maintained company profile with raw source text. Creating or updating memory alone leaves a draft; activation is the explicit publish step.

Recruit and configure an employee

  1. Call list_role_templates.
  2. Use create_agent with a clear display name. New employees start with every workspace tool enabled; disable what the role does not need.
  3. Use write_agent_config for role-specific operating guidance.
  4. Set an available model and required skills.
  5. Call provision_agent.
  6. Confirm the employee with list_agents and runtime health.
  7. Create one known test task before enabling more tools.

Configuration and provisioning require an owner or admin role.

Install a custom employee skill or helper script

  1. Identify the employee with list_agents.
  2. Review the file contents and choose a relative path beneath skills/ or scripts/.
  3. Call write_agent_file as an owner or admin.
  4. Provision the employee if needed, then run a bounded test task.
  5. Inspect the resulting task, messages, and files before widening the employee's tool surface.

write_agent_file cannot replace runtime-owned root documents, write .env, use an absolute path, or traverse outside the employee workspace. It returns file metadata rather than echoing the file body.

Rotate a workspace or employee secret

  1. Call list_secrets to inspect names, scope, provider, and update time without exposing values.
  2. Decide whether the new value should be shared workspace-wide or override that name for one employee.
  3. Call set_workspace_secret; include agentId only for the employee-scoped override.
  4. Confirm the returned scope and version metadata.
  5. Run a bounded employee test and revoke the old credential at its source when appropriate.

Secret values are write-only. A same-named employee secret takes precedence over the workspace value for that employee.

Register a private MCP server

Use add_custom_mcp_server only after reviewing the remote server's tool surface. It requires secrets:write because an optional authorization header is encrypted as a workspace secret. The tool returns server metadata, never the header value.

After registration, verify and grant the server through Yodu's employee tool-access workflow. Platform MCP registration does not automatically make every external action appropriate for every employee.

On this page