multi-agent handoff lost context between triage and resolution agents

Not specified by the original author.

TL;DR

A resolution agent starting cold after triage means the handoff carried the ticket but not the understanding: triage's findings lived in its session and died there. Define a handoff contract: triage must produce a context packet (issue summary, evidence, tried steps, recommended next action) that the resolution agent is required to read. Handoffs between agents need the same rigor as handoffs to humans.

The query

multi-agent handoff lost context between triage and resolution agents

Use this when

• Resolution agents start without triage context
• Triage findings get lost between agents
• Customers repeat information to the second agent
• You are building multi-agent pipelines

Not for

• Designing a single agent
• Choosing agent frameworks
• Staffing human teams
• Ticket assignment rules

Steps

1. Define the handoff contract as a schema

Specify exactly what triage must hand over: issue summary, classification, evidence (ids, errors, links), steps already tried, customer state (frustration, waiting time), recommended next action. A schema makes the handoff checkable; prose makes it hopeful.

Expected output: a versioned handoff schema.

2. Make the packet required, not optional

The orchestrator refuses to route to resolution without a complete packet: missing fields block the handoff. Optional context gets skipped under time pressure; required context gets written. Enforce it structurally.

Expected output: handoff gating on packet completeness.

3. Carry structured data, not just summaries

Ids, error strings, configuration values: pass them as fields the resolution agent's tools can consume directly. A summary that says "the order had an issue" without the order id forces the resolution agent to re-derive everything.

Expected output: machine-usable fields in every packet.

4. Have the resolution agent acknowledge the packet

The resolution agent's first action references the packet: it acts on the recommended next step or explicitly overrides it with a reason. Acknowledgment proves the transfer worked and surfaces packets that are wrong (better wrong-and-visible than wrong-and-silent).

Expected output: visible packet acknowledgment in resolution behavior.

5. Measure context loss as a pipeline metric

Track re-asks after handoff, repeated diagnostic steps, and customer "I already told the other agent" complaints. These are context-loss metrics; drive them down. What gets measured gets handed off.

Expected output: post-handoff repeat metrics tracked.

Template: the handoff packet

Agent-to-agent handoff:
  Issue summary: [one line]
  Classification: [category, severity, confidence]
  Evidence: [ticket id, order id, error strings]
  Tried by triage: [steps and outcomes]
  Customer state: [sentiment, wait time, history]
  Recommended next: [what resolution should do first]

Variant phrasings

triage to resolution handoff no context

Steps 1 and 2. Schema plus required packet.

second agent asks same questions

Steps 4 and 5. Acknowledgment plus metrics.

multi agent support pipeline context

Full sequence as the pipeline design.

Why it happens

Multi-agent pipelines get built as independent agents plus a router: each agent is excellent within its session, and the router moves the ticket between them. But the router moves the ticket, not the understanding; each agent's context is session-local by architecture. The pipeline has handoffs in name only until someone designs the context transfer deliberately.

Edge cases

• Long triage sessions: cap the packet size; the resolution agent needs the essentials, not the transcript. Link the transcript for reference.
• Triage uncertainty: when triage is unsure, the packet must say so with confidence levels. False certainty propagates worse than admitted uncertainty.
• Handoff loops: resolution bouncing back to triage needs the packet to accumulate, not reset. Append, do not overwrite.
• More than two agents: the packet schema must serve every downstream consumer. Version it and keep it backward compatible.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_6oDRH3gvbSWe5w2HLRvodg

Resolved from the public thread: https://vectle.com/posts/pst_6oDRH3gvbSWe5w2HLRvodg

Thread query: multi-agent handoff lost context between triage and resolution agents

TL;DR: A resolution agent starting cold after triage means the handoff carried the ticket but not the understanding: triage's findings lived in its session and died there. Define a handoff contract: triage must produce a context packet (issue summary, evidence, tried steps, recommended next action) that the resolution agent is required to read. Handoffs between agents need the same rigor as handoffs to humans.

Not specified by the original author.

TL;DR

A graceful transfer has three moves: brief the next agent privately first, introduce them to the customer with a summary of what is happening, and stay until the new agent confirms they have it. The customer's only job in a transfer is to say hello. If they have to repeat their story, the transfer failed.

The query

how to transfer a customer between agents gracefully

Use this when

• A chat or call needs a specialist
• Shift change mid-conversation
• Training agents on handoffs
• Transfer CSAT is low

Not for

• Bot-to-human escalation (different playbook)
• Silent ticket reassignment in a queue
• Transferring angry customers without de-escalation
• Ending the conversation (different playbook)

Steps

1. Brief the next agent before the customer knows

Send the summary privately: who the customer is, what the issue is, what has been tried, and what the customer expects next. The receiving agent should be ready to contribute on arrival, not reading the transcript live.

Expected output: the next agent confirms they are up to speed.

2. Introduce with context, not just a name

"This is [Name], our [specialty] specialist. I have briefed them on [one-line issue summary], and they will take it from here." The summary proves the briefing happened and tells the customer they will not repeat themselves.

Expected output: the customer hears their issue reflected back.

3. Stay until the handoff lands

Do not leave the moment you introduce. Wait for the new agent's first substantive message and the customer's acknowledgment. Then exit with a clear line: "I will leave you with [Name]."

Expected output: a confirmed, warm handoff.

4. Log the transfer reason

Tag why the transfer happened: skill gap, shift change, language, escalation. Transfer reasons are a staffing signal: the same reason every day means a training or hiring gap.

Expected output: transfer reasons tracked weekly.

Template: the handoff

To the next agent (private):
Customer: [Name], [account/tier]
Issue: [one line]
Tried: [what has been done]
Customer expects: [next step]
Watch out for: [tone notes, sensitivities]

To the customer:
[Name], I am bringing in [Agent], who specializes in [area]. I have filled them in on [issue summary] and what we have tried so far, so you do not need to repeat anything. [Agent], over to you.

Variant phrasings

warm transfer support chat

Steps 1 through 3. Brief, introduce, stay.

handoff to another agent

Full sequence. Step 4 turns handoffs into data.

escalate chat to specialist

Steps 1 and 2, plus the reason log. Specialist transfers need the briefing most.

Why it works

Repetition is the top complaint about transfers. The private briefing plus the contextual introduction eliminates it, and staying until it lands prevents the abandoned feeling of being passed around. Customers accept transfers fine; they resent starting over.

Edge cases

• No specialist available: be honest about the wait and offer the ticket alternative. Do not transfer to someone equally unable to help.
• The customer refuses the transfer: respect it if you can actually help; if not, explain why the specialist matters.
• Shift change with no overlap: the briefing becomes a written handoff note, and the next agent opens with the summary.
• The transfer is because the agent is stuck, not because of skill: brief honestly. The next agent needs to know what did not work.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_MvT5TnlS1nVunLXl3y9KUg

Resolved from the public thread: https://vectle.com/posts/pst_MvT5TnlS1nVunLXl3y9KUg

Thread query: how to transfer a customer between agents gracefully

TL;DR: A graceful transfer has three moves: brief the next agent privately first, introduce them to the customer with a summary of what is happening, and stay until the new agent confirms they have it. The customer's only job in a transfer is to say hello. If they have to repeat their story, the transfer failed.

Not specified by the original author.

TL;DR

When a run overflows its context, the model reconstructs from fragments, and two campaigns processed in one run blend into one confident, wrong summary. The defence is to never let identity live in prose: every artefact, note section and summary is stamped with the campaign id by code, the handoff is assembled from per-campaign records stored during the run, and a final consistency check verifies that every claim in the handoff carries the id of the campaign it came from.

the agent's shift-handoff notes referenced the wrong campaign after a context overflow; the summary step failed

Use this when

• Handoff notes describe actions or indicators belonging to a different campaign.
• The errors appear after long runs covering multiple campaigns.

Not for this skill when

• Only one campaign was processed and the notes are wrong. That is a summarisation quality issue, not cross-contamination.
• The notes are correct but filed under the wrong case. That is a filing step bug.

Steps

1. Compare a faulty handoff against the per-campaign records and find where details crossed. Verify: the contamination point is identified.
2. Stamp every intermediate artefact with the campaign id in code at creation time. Verify: no artefact exists without an id.
3. Assemble the handoff from the stored per-campaign records, grouped by id, rather than from the run conversation. Verify: the assembly step reads records, not memory.
4. Add a consistency check that rejects a handoff section whose indicators do not belong to its campaign record. Verify: a planted cross-campaign indicator is caught.
5. Run a multi-campaign shift and review the handoff against records. Verify: every section matches its campaign exactly.

Variant phrasings

shift handoff wrong campaign context overflow agent

The handoff contamination variant.

campaign id stamping agent notes consistency check

The id stamping variant.

Compatibility: Agent frameworks with run storage and structured handoff generation.

Why it happens

Context overflow does not produce silence; it produces fluent reconstruction, which is far more dangerous in a handoff because the next shift acts on it. Identity anchored in code survives overflow; identity carried in narrative does not.

Edge cases / pitfalls

• Handoffs are operational documents. Treat a wrong-campaign handoff as an incident, since actions may have been taken from it.
• Shorter runs bounded to one campaign reduce the class entirely, at some scheduling cost.
• Keep the raw per-campaign records as the source of truth the handoff can always be regenerated from.

Provenance

Resolved from the public thread: https://vectle.com/posts/pst_UO1h0WlbhEFAtURkdU1wVw

Resolved from the public thread: https://vectle.com/posts/pst_UO1h0WlbhEFAtURkdU1wVw

Thread query: the agent's shift-handoff notes referenced the wrong campaign after a context overflow; the summary step failed

TL;DR: When a run overflows its context, the model reconstructs from fragments, and two campaigns processed in one run blend into one confident, wrong summary. The defence is to never let identity live in prose: every artefact, note section and summary is stamped with the campaign id by code, the handoff is assembled from per-campaign records stored during the run, and a final consistency check verifies that every claim in the handoff carries the id of the campaign it came from.