--- schema: formation.pattern/v0.1 kind: pattern visibility: public aliases: - 'context handoff' - 'session handover' - 'relay agents' - 'long-running agent continuation' canonical_url: https://topologyindex.com/patterns/successor_handoff.md description: 'An agent hands its accumulated context to a later agent that continues the work.' executable_here: false executor_requirements: - context_carried_between_episodes family: continuity findings_page: /patterns/successor_handoff/findings.md observed_in: - /research/openai-hugging-face-agent-coordination.md path: /patterns/successor_handoff.md pattern: successor_handoff pattern_index: /patterns/index.md product_api_version: v1 published_findings: - benchmarks: [] compared_against: 'A coding agent looping across context windows with compaction only' direction: helped source_id: web:anthropic.com/engineering/effective-harnesses-for-long-running-agents task_domain: 'Long-running software development' task_domains: - coding url: https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents - benchmarks: [] compared_against: 'The same agent without context management features' direction: helped source_id: web:claude.com/blog/context-management task_domain: 'Agentic web search' task_domains: - research url: https://claude.com/blog/context-management - benchmarks: [] compared_against: 'Keeping all history in one context window' direction: mixed source_id: web:anthropic.com/engineering/effective-context-engineering-for-ai-agents task_domain: 'Long-horizon agent tasks' task_domains: [] url: https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents - benchmarks: - 'Deep Memory Retrieval' compared_against: 'Fixed-context LLMs without memory management' direction: helped source_id: arxiv:2310.08560 task_domain: 'Multi-session chat and long document analysis' task_domains: - documents url: https://arxiv.org/abs/2310.08560 - benchmarks: [] compared_against: 'Information spread across many conversation turns' direction: helped source_id: arxiv:2505.06120 task_domain: 'Multi-turn generation tasks' task_domains: [] url: https://arxiv.org/abs/2505.06120 qualifying_evidence: [] references: [] schema_version: v0.1 title: 'When to use the successor_handoff multi-agent pattern (context handoff, session handover)?' unmet_executor_requirements: - context_carried_between_episodes --- # When to use the successor_handoff multi-agent pattern (context handoff, session handover)? **Short answer:** start with it when the task outlasts one context window or session; avoid it when rediscovery is cheaper than a handover. A hypothesis from the [decision guide](/patterns/index.md), not a ranking; [published findings](/patterns/successor_handoff/findings.md) keep the unfavourable ones. The `successor_handoff` multi-agent pattern (family `continuity`): an agent hands its accumulated context to a later agent that continues the work. Also called: context handoff, session handover, relay agents, long-running agent continuation. ## The arrangement - Structure: A predecessor and a successor that never run at the same time. - Control: The handover document is the control: what was done, what to do next. - Shared state: The handed-over context, and nothing else. ## When it is hypothesised to fit Hypothesised to fit when work outlasts one context window or one sandbox, and rediscovery costs more than writing the handover. A hypothesis to test on a workload, not a finding: this service has not tested it. ## Failure modes to watch for - A handover is a summary: it carries the predecessor’s mistakes and drops what the predecessor thought unimportant. - The successor inherits confidence it did not earn, and usually cannot tell which claims were verified. - Nothing in a handover separates a fact from an assumption unless the format forces it. These are things to watch for, not outcomes anyone measured here. ## What published studies found Attributed to each source and stated without figures; unfavourable results are included on purpose, and none of this is evidence produced by this service. What each was compared with is in `published_findings`; each finding in words, with its caveat, is at [/patterns/successor_handoff/findings.md](/patterns/successor_handoff/findings.md). Reviewed 2026-09-23. Each label says how this pattern fared against what it was compared with, as the source reports it; the labels are defined at [/docs/schemas/pattern/v0.1.md](/docs/schemas/pattern/v0.1.md). - **Helped**: [Anthropic Engineering: Effective harnesses for long-running agents](https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents) - **Helped**: [Claude blog: Managing context on the Claude Developer Platform](https://claude.com/blog/context-management) - **Mixed**: [Anthropic Engineering: Effective context engineering for AI agents](https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents) - **Helped**: [MemGPT: Towards LLMs as Operating Systems](https://arxiv.org/abs/2310.08560) - **Helped**: [LLMs Get Lost In Multi-Turn Conversation](https://arxiv.org/abs/2505.06120) ## Can this deployment execute it No. The runner does not provide everything this arrangement needs. What is missing: - `context_carried_between_episodes` (what one episode learned carried into a later one): Every episode starts in a fresh isolated workspace, which is also what keeps one episode from advantaging the next. A topology formation that declares what this deployment cannot run is still stored, and is refused with an unsupported-capability error rather than reduced to a simpler arrangement and run anyway. ## Evidence for this pattern None. No qualifying evidence exists here for this pattern, and by design none ever will: an evidence snapshot is about the complete execution configuration that produced the observations, so the most it can support is a claim about that configuration on that workload. ## Observed in Reports that describe this arrangement. A report is a reason to test an arrangement, never a result for it. - [Agent coordination in the OpenAI and Hugging Face evaluation incident](/research/openai-hugging-face-agent-coordination.md) (descriptive_only) ## How evidence works here A pattern names how work is organized, never how well that works. Evidence in this service attaches to a complete execution configuration — a topology formation together with an execution lock pinning models, tools and runtime — and never to a pattern on its own, so no page here carries a success rate, a cost or a run count for an arrangement.