Common Agent patterns

Use these patterns as starting points, then test with your own data and failure cases.

Ongoing research assistant

• Why: repeated research benefits from a continuing workspace.
• Configuration: a citation-focused prompt, relevant research Skills, optional Remote HTTPS MCP sources, and a reusable Environment for analysis packages.
• Delivery: mosoo Threads for interactive work; API access for embedding in another product.

Pull-request reviewer

• Why: each review should start in clean temporary state.
• Configuration: repository-review Skill, GitHub MCP connection, strict output format, and a minimal Environment.
• Delivery: one Thread per review or an external system calling the Public Thread API.

Ticket triage worker

• Why: tickets are independent jobs and should not leak temporary state.
• Configuration: triage rubric in a Skill, issue-tracker MCP, and explicit rules for labels and escalation.
• Operations: filter failed Threads, inspect Logs, and monitor production usage separately from Preview.

Team copilot

• Why: the copilot benefits from a stable working directory across follow-up turns in the same Session.
• Configuration: product and operations Skills plus authorized MCP connections.
• Boundary: mosoo's current console is single-owner; team roles and shared administration are not available. Expose the Agent through an appropriate external surface rather than sharing owner credentials.

Public application

Use a Project key from your backend to create or continue a Session for each application user. Supply harness configuration directly or reuse an optional Agent preset. Existing v1 integrations can keep their published-Agent flow.