Handover field guide
Build AI work that another actor can continue
Practical, source-linked guides for agent handoffs, shared context, MCP workflows, service identities, and review of AI-generated work.
MCP shared context
How to hand off shared context between AI agents with MCP
Use an MCP server and a durable continuation record to move objectives, state, decisions, evidence, review, and next actions between AI agents.
Read guideContext engineering
Context engineering for AI agents: memory vs shared context
Design context for AI agents across prompts, runtime memory, shared records, and source artifacts without turning every continuity problem into vector search.
Read guideArchitecture decision
AI agent handoff vs memory vs routing
Compare agent routing, runtime memory, and durable handoffs, then choose the right continuity layer for each boundary in a production agent system.
Read guideArchitecture decision
Handover vs HANDOFF.md: when is a file enough?
Use the File Sufficiency Test to decide when a HANDOFF.md file is enough for agent continuity and when a versioned, access-controlled handover is required.
Read guideHuman review
Human-in-the-loop review for AI agent output
Use revision-anchored annotations, named reviewers, and verifiable corrections to close the feedback loop between humans and AI agents.
Read guideRunnable MCP workflow
End-to-end MCP agent handoff: publish, review, correct, and continue
Run a complete MCP handoff across separate identities with Markdown, SQL, JSON, revision-anchored review, a verified correction, and a successor continuation.
Read guideMCP workflow
A practical MCP workflow for multi-agent collaboration
Connect several AI agents to one durable handoff record, preserve authorship, and continue work safely across tools with MCP.
Read guideCross-model workflow
Move context between Claude, ChatGPT, Gemini, Cursor, and Codex
Move active work between leading AI hosts without relying on one provider's chat history, memory format, or hidden state.
Read guideHandoff record
What belongs in an AI agent handoff record?
Use a concrete seven-part record to preserve the task state another AI agent or person needs to verify and continue the work.
Read guideCompany knowledge
A shared knowledge base for AI agents and human teams
Design a company knowledge layer where people and AI agents can retrieve approved sources, publish attributable work, and continue it without crossing access boundaries.
Read guideAgent identity
Service accounts for AI agents
Give unattended AI agents attributable, revocable access without sharing a person's login or collapsing several agents into one identity.
Read guideMCP authentication
MCP OAuth vs service accounts for AI agents
Choose OAuth for an interactive person and a scoped service identity for unattended agents without confusing authentication, authorship, or access.
Read guideVersioned reports
How to version and review AI-generated reports
Preserve each generated report, its source evidence, review comments, and follow-up revisions as one inspectable continuation record.
Read guideContext architecture
MCP context server vs vector memory
Compare MCP context servers with vector memory, learn what each layer does, and choose an architecture for searchable, reviewable AI work.
Read guideCoding-agent continuity
How to preserve context across AI coding agents
Use repository instructions, a changing handoff record, Git state, and a shared revision to move unfinished coding work between Codex, Claude Code, Cursor, and Gemini CLI.
Read guideClaude Code to Codex
How to transfer context from Claude Code to Codex
Use Codex import for a one-time Claude Code migration, or use a versioned Handover record when the two agents need to exchange unfinished work repeatedly.
Read guideMCP setup
Use an MCP memory server with Claude Code, Cursor, and Codex
Connect Claude Code, Cursor, and Codex to one remote MCP context server, verify the authenticated identity, and test a complete publish-and-retrieve handoff.
Read guideReport migration
How to migrate static report folders into shared AI context
Map products, folders, HTML reports, supporting files, URLs, and publishing workflows into versioned Spaces and handovers without losing provenance or forcing an immediate cutover.
Read guideAgent governance
AI agent governance framework: controls, checklist, and tests
Use 18 practical controls for AI agent identity, access, approvals, evidence, testing, revocation, and offboarding, plus a private interactive checklist.
Read guideAgent Skills + MCP
How Agent Skills and MCP work together
Use a portable SKILL.md to teach an agent a reliable workflow, then use MCP to provide authenticated tools and data without putting credentials or authorization decisions in the skill.
Read guideCross-agent installation
Install Agent Skills in Claude Code, Codex, Cursor, and Gemini CLI
Install one open Agent Skill across four coding-agent hosts, verify that each host discovers it, connect the authenticated MCP runtime, and prove the workflow completed.
Read guideConnection verification
How to test an MCP server connection from the CLI
Verify an MCP-backed context service from the command line by checking its endpoint, authenticated identity, access scope, and a real context read.
Read guideInstallable workflow
AI agent handoff skill for durable context
Install an open Agent Skill that teaches Codex, Claude Code, Cursor, and Gemini CLI to create, verify, retrieve, review, and continue durable AI agent handoffs.
Read guideCline integration
Cline MCP server setup for durable agent context
Connect Cline IDE or Cline CLI to a remote Streamable HTTP MCP server, verify the authenticated identity, and test a cross-agent context handoff.
Read guideGemini CLI integration
Gemini CLI MCP server setup for shared agent context
Add a remote Streamable HTTP MCP server to Gemini CLI, diagnose disconnected and authentication states, and prove a second agent can continue the exact work.
Read guideClaude Code integration
Claude Code MCP server setup for durable agent context
Add a remote HTTP MCP server to Claude Code, diagnose authentication and policy states, and prove a fresh agent can retrieve and continue exact shared context.
Read guideCursor integration
Cursor MCP server setup for durable agent context
Configure a remote MCP server in Cursor IDE or Agent CLI, diagnose connection and OAuth states, and prove a fresh agent can retrieve and continue exact shared context.
Read guideCodex integration
Codex MCP server setup for durable agent context
Add Handover as a remote MCP server in Codex CLI, the IDE extension, or the desktop app, verify the authenticated boundary, and prove a fresh agent can continue exact shared context.
Read guideCompany operations
Organizational context handoffs for human and AI teams
Transfer accountable work between teams, people, and AI agents with explicit scope, access, evidence, receiver acceptance, and rollback.
Read guide