# Agent Handoff Prose Capacity

<!-- auto-oncall-optimization-record
{
  "schemaVersion": "auto-oncall-optimization-record-v1",
  "optimizationId": "agent-handoff-prose-capacity-2026-08-12",
  "problems": [
    {
      "problemId": "agent-handoff-character-cap-cuts-diagnostic-sentence",
      "summary": "The 500-character evidence question and stop-condition schemas forced incomplete sentences, while separate 6000 and 1200 character final wording caps could discard load-bearing diagnosis before the real publication byte boundary.",
      "status": "optimized_verified",
      "issueKeys": ["APID-29296"],
      "sourceRefs": ["https://tidb.atlassian.net/browse/APID-29296#problem:agent-handoff-text-truncation"],
      "signalTextIncludesAny": ["500 character evidence question", "question truncated mid-sentence", "agent handoff text too large"],
      "observedThrough": "2026-08-12",
      "effectiveAt": "2026-08-12T11:13:00+08:00",
      "verifiedAt": "2026-08-12T11:40:00+08:00",
      "attemptedAction": "Remove small semantic character caps from evidence questions, refinement stop conditions, and Router-to-Final wording while retaining artifact IDs, compact evidence byte budgets, batch cardinality, and the configured Jira publication byte boundary.",
      "evidenceUrl": "https://github.com/tidbcloud/aisupport/tree/main/auto-oncall-services/docs/opts/2026-08-12-agent-handoff-prose-capacity.md"
    }
  ]
}
-->

- Owner: Scenario Diagnostic Agent, Scenario Router, and Final Response Agent
- Source: `APID-29296` replay `c530fc32-0448-474a-b21b-79f1b019c9ac`
- Status: optimized_verified

Prompts still require concise questions and answers, but semantic prose is not
cut at a small character boundary. The 16 KiB fact and 128 KiB scenario handoff
remain explicit compact-evidence boundaries; Jira publication continues to use
the configured byte limit.

Staging replay `d8c9f1ba-e9ae-42b5-b338-e190ccbd27d9` preserved both initial
Evidence questions and their stop conditions beyond the former 500-character
boundary. Accumulated scenario handoffs reached 64,530 and 45,849 bytes without
the former 1,000-character truncation, and Router-to-Final content remained
complete.

The existing Jira internal comment `23399342` was updated in place after human
review. Readback confirmed `public=false` and the expected body; no duplicate
comment or issue transition was created.

Next action: monitor later multi-round handoffs for semantic sentence loss.