eval(pkm): use framework transcript recording

[sparkleMing]

Summary

• upgrade to so evals use the framework transcript recorder and EventBus close handling
• pass the eval into while preserving the default production controller path
• remove PKM harness-owned trajectory reconstruction from outcomes; keep outcomes focused on final state and
• update PKM graders to read tool order and tool payloads from
• add focused grader regression tests for workspace diff routing and read-before-write transcript ordering

Validation

• Analyzing 2 items...
  No issues found! (ran in 5.6s)
• 00:00 +0: loading /Users/ming/Downloads/project/opensource/memex/test/agent/eval/pkm_agent/graders_test.dart
  00:00 +0: PKM eval graders routes from workspace diff, not outcome trajectory fields
  00:00 +1: PKM eval graders read-before-write uses transcript tool-call order
  00:00 +2: All tests passed!
• Analyzing 4 items...
  No issues found! (ran in 3.9s)

Did not run the full live PKM suite because runs the whole suite from a single test entrypoint.

[github-actions]

PR AI Review / PR AI 语义预检

中文

• 风险等级：低风险
• 需要人工审核：否
• 黄金链路影响：无
• 置信度：medium
• Workflow run：26578118635

此 PR 重构 PKM eval 测试框架：升级 dart_agent_core 至 1.0.13，使 eval 使用框架内置的 transcript 录制；PkmAgent.runWithContent 新增可选 controller 参数供 eval 注入；移除 harness 中手动重建工具调用历史的代码，改为从 outcome workspace diff 和 transcript 读取数据；新增两个 grader 回归测试。生产代码变更极小（仅新增可选参数），风险低。

影响范围

• agent
• tests

黄金链路

• 未识别到黄金链路影响。
• 说明：变更仅涉及 eval 测试基础设施和 dart_agent_core 小版本升级。生产代码中新增的 controller 参数为可选且向后兼容，不影响任何核心用户链路。

风险项

• info 移除 harness 中手动重建工具历史的代码，改用框架 transcript。证据：test/agent/eval/pkm_agent/harness.dart: removed ~80 lines of _replayToolHistoryFromState and replaced with _diffSnapshots, test/agent/eval/pkm_agent/graders.dart: graders now read from transcript.toolCalls and outcome.workspaceDiff。
  建议：变更合理，减少了 harness 与 agent 内部状态的耦合。无需额外操作。
• info PkmAgent 新增可选 controller 参数，向后兼容。证据：lib/agent/pkm_agent/pkm_agent.dart: AgentController? controller added to createAgent() and runWithContent(), Default path preserved: final agentController = controller ?? AgentController()。
  建议：参数为可选且默认值不变，现有调用方无需修改。无架构违规。

测试缺口

• agent 新增的可选 controller 参数在生产路径中没有直接的单元测试覆盖（仅通过 eval 集成测试间接验证）。 建议检查：可接受，因为 eval 测试覆盖了该路径；但如果未来有更多调用方使用此参数，建议补充单测。。

English

• Risk level: LOW
• Human review required: NO
• Golden path impact: NONE
• Confidence: medium
• Workflow run: 26578118635

This PR refactors the PKM eval harness: bumps dart_agent_core to 1.0.13 so evals use the framework transcript recorder; adds an optional controller parameter to PkmAgent.runWithContent/createAgent for eval injection; removes harness-owned trajectory reconstruction, switching graders to read from outcome workspace diff and transcript; adds two grader regression tests. Production code change is minimal (one optional parameter added), risk is low.

Affected Areas

• agent
• tests

Golden Path

• No golden path impact was identified.
• Rationale: Changes are limited to eval test infrastructure and a minor dart_agent_core version bump. The new optional controller parameter in production code is backward-compatible and does not affect any core user flow.

Findings

• info Removed harness-owned trajectory reconstruction in favor of framework transcript. Evidence: test/agent/eval/pkm_agent/harness.dart: removed ~80 lines of _replayToolHistoryFromState and replaced with _diffSnapshots, test/agent/eval/pkm_agent/graders.dart: graders now read from transcript.toolCalls and outcome.workspaceDiff.
  Recommendation: Change is sound — reduces coupling between harness and agent internal state. No action needed.
• info PkmAgent gains optional controller parameter, backward-compatible. Evidence: lib/agent/pkm_agent/pkm_agent.dart: AgentController? controller added to createAgent() and runWithContent(), Default path preserved: final agentController = controller ?? AgentController().
  Recommendation: Parameter is optional with unchanged default; existing callers are unaffected. No architecture violation.

Test Gaps

• agent The new optional controller parameter has no direct unit test for the production code path (only indirectly validated via eval integration tests). Suggested check: 可接受，因为 eval 测试覆盖了该路径；但如果未来有更多调用方使用此参数，建议补充单测。.

AI review is advisory. Maintainers should verify the result before merging.

[github-actions]

PR Preflight Summary / PR 预检汇总

中文

• 统一结论：低风险：两个预检均已完成，质量预检通过，可走普通手动合并流程。
• Policy preflight：低风险。未命中打回、高风险或警告规则。
• Flutter quality：通过。Analyzer 和 test baseline 均未发现新增问题。
• PR head：4a8ef445f13ece62207c3079485a4006f3b96aec
• Policy run：26578118633
• Flutter run：26578119137

English

• Combined result: Low risk: both preflights completed and quality passed; use the normal manual merge flow.
• Policy preflight: LOW RISK. No blocking, high-risk, or warning policy signal was found.
• Flutter quality: PASS. Analyzer and test baselines found no newly introduced issue.
• PR head: 4a8ef445f13ece62207c3079485a4006f3b96aec
• Policy run: 26578118633
• Flutter run: 26578119137

PR Policy Preflight / PR 规则预检

PR Policy Preflight / PR 规则预检

中文

• 判定：低风险
• 变更文件数：7
• 变更行数：499
• Diff 是否截断：false

未发现确定性规则问题。

English

• Decision: LOW RISK
• Changed files: 7
• Changed lines: 499
• Diff truncated: false

No deterministic policy findings.

PR Flutter Quality / Flutter 质量预检

PR Flutter Quality / Flutter 质量预检

中文

• 总体：通过
• Analyzer baseline：通过
• Test baseline：通过

English

• Overall: PASS
• Analyzer baseline: PASS
• Test baseline: PASS

Flutter Analyzer Baseline

• Base issues: 307
• PR issues: 307
• New issues: 0

No new analyzer issues introduced by this PR.

Flutter Test Baseline

• Base failures: 0
• PR failures: 0
• New failures: 0

No new Flutter test failures introduced by this PR.