hmharness 自进化证据

本页由 scripts/export-evidence.cjs 从 HMH_HOME 的真实日志生成,数字只来自文件,缺席标注为 absent。协议见 docs/SELFFEED.md;原始数据在 raw/。

窗口:2026-08-26T17:42:17.903Z → 2026-09-12T03:24:49.765Z · 生成于 2026-09-12T03:30:38.695Z

⚠ 外推性声明:当前数据全部来自自设任务(scripts/selffeed-tasks.json,作者编写),bench 用例同源。进化系统在此人造适应度景观上的表现不代表其在真实外部开发者任务上的效果。外推性待第一批外部用户数据验证。

每日进化轮次

日期	轮次
2026-08-26	2
2026-08-27	1
2026-08-29	1
2026-09-02	2
2026-09-03	1
2026-09-04	2
2026-09-05	1
2026-09-07	11
2026-09-08	22
2026-09-09	13
2026-09-10	12
2026-09-11	10
2026-09-12	3

技能判例(晋升/拒绝,含理由——诚实优先)

日期	技能	判定	理由
2026-09-12	harmony-api-lookup-blindspot-workaround	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	continue-hmharness-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	windows-installer-hash-signature-verification-fallback	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-failure-environment-triage	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-vision-provider-capability-marking	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-emulator-ui-regression-loop	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	windows-self-media-publishing-toolbox	promoted	no regression (train 50% vs 50%, holdout 0%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	windows-large-vision-request-payload-file	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-vision-false-negative-triage	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-vision-ui-regression-fallback	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-vision-provider-routing-fix	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-11	harmony-uitest-layout-ground-truth	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	windows-browser-publish-preflight	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	browser-automation-preflight	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	read-verify-file-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-10	harmony-worktree-runtime-reverify	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	harmony-runtime-log-verification	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	harmony-device-screenshot-verification	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	read-verify-file-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-10	read-verify-file-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-10	harmony-worktree-resume-reverify	promoted	no regression (train 50% vs 50%, holdout 0%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	serialize-same-file-edits	promoted	no regression (train 50% vs 50%, holdout 0%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	read-verify-file-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-10	harmony-device-test-explicit-bundle	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	truncated-verify-path-resolution	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-10	turn-budget-aware-verification	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	network-preflight-for-harmony-tools	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	windows-file-path-verification	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	read-verify-file-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	arkui-scroll-verify-pattern	promoted	no regression (train 100% vs 50%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	verify-task-turn-budget-gating	promoted	no regression (train 100% vs 50%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	continue-hmharness-conversation-workflow	promoted	no regression (train 50% vs 50%, holdout 0%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	windows-worktree-path-preflight	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	verify-staged-task-runner	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	codexhost-bridge-stuck-thinking	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	harmony-staged-task-executor	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	harmony-staged-verify-executor	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	harmony-md-task-executor	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-09	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-08	codexhost-upstream-merge-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	hmh-noninteractive-execution-in-codexhost	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	codexhost-streaming-bridge	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	hmh-noninteractive-execution	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	codexhost-source-merge-upgrade	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	codexhost-offline-upgrade	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	verify-before-execution-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-08	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-08	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-08	harmony-sdk-symbol-verification	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	windows-harmony-toolchain-path-resolution	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	continue-hmharness-conversation-workflow	promoted	no regression (train 50% vs 50%, holdout 0%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-08	windows-curl-download-retry	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion)
2026-09-08	continue-hmharness-conversation-workflow	promoted	no regression (train 50% vs 50%, holdout 0%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived
2026-09-08	continue-hmharness-conversation-workflow	promoted	no regression (train 100% vs 100%, holdout 100%) - promoted to CANARY (20% sessions, impact-gated full promotion); previous version archived

被拒候选存档(Pareto 池)

日期	候选	拒因
no rejected candidates archived yet

最近洞察

日期	洞察
2026-09-12	{"time":"2026-09-12T03:25:11.084Z","session":"2026-09-12T03-16-55-yxrvhk","task":"继续检查进度和推进","outcome":"ok","turns":15,"toolUses":14,"toolsUsed":["run_command",
2026-09-12	{"time":"2026-09-12T03:19:24.871Z","session":"2026-09-12T03-18-20-ta45j3","task":"Look up SDK symbols for UIAbility lifecycle and Want via harmony_api_lookup; v
2026-09-12	{"time":"2026-09-12T03:00:22.499Z","session":"2026-09-12T02-59-31-z8r77k","task":"按你的计划推进","outcome":"ok","turns":2,"toolUses":2,"toolsUsed":["run_command","rem
2026-09-12	{"time":"2026-09-12T02:59:55.213Z","session":"2026-09-12T02-59-39-qp4tn6","task":"HMH\\_CODEXHOST\\_070\\_20260912\\_OKHMH\\_CODEXHOST\\_070\\_20260912\\_OK","o
2026-09-12	{"time":"2026-09-12T02:09:08.140Z","session":"2026-09-12T02-02-40-tjng9t","task":"继续","outcome":"ok","turns":4,"toolUses":3,"toolsUsed":["write_file","run_comma
2026-09-11	{"time":"2026-09-11T19:01:41.198Z","session":"2026-09-11T18-59-45-5uenou","task":"继续","outcome":"ok","turns":1,"toolUses":0,"toolsUsed":[],"skillsInjected":["br
2026-09-11	{"time":"2026-09-11T18:19:50.291Z","session":"2026-09-11T18-19-23-j0osts","task":"需要，装Ubuntu最新版本和一个Windows 11 客户机，如果可以，也装Kali linux.记得都要中文版本的","outcome":"ok","t
2026-09-11	{"time":"2026-09-11T18:16:35.426Z","session":"2026-09-11T18-16-02-rffuc0","task":"记得装完系统先装 VMware Tools（虚拟机菜单 → 安装 VMware Tools），它提供显卡驱动、剪贴板共享、自适应分辨率——Windows 客

技能清单

active:

• browser-harness
• cangjie-win-static-pitfalls
• cjpm-build-repair
• computer-use
• delegate-task
• graphify
• gsap-core
• gsap-frameworks
• gsap-performance
• gsap-plugins
• gsap-react
• gsap-scrolltrigger
• gsap-skills
• gsap-timeline
• gsap-utils
• hdc-device-ops
• skill-discovery
• video-shotcraft
• vpn-manager
• browser-automation-preflight
• file-op-error-recovery
• harmony-device-screenshot-verification
• harmony-ops-flow
• harmony-runtime-log-verification
• harmony-toolchain-report
• harmony-worktree-runtime-reverify
• network-preflight-for-harmony-tools
• safe-file-ops
• safe-path-existence-check
• turn-budget-aware-verification
• turn-budget-checkpointing
• windows-command-compatibility
• windows-file-path-verification

canary:

• continue-hmharness-workflow
• harmony-api-lookup-blindspot-workaround
• harmony-device-test-explicit-bundle
• harmony-emulator-ui-regression-loop
• harmony-failure-environment-triage
• harmony-uitest-layout-ground-truth
• harmony-vision-false-negative-triage
• harmony-vision-provider-capability-marking
• harmony-vision-provider-routing-fix
• harmony-vision-ui-regression-fallback
• harmony-worktree-resume-reverify
• serialize-same-file-edits
• truncated-verify-path-resolution
• windows-browser-publish-preflight
• windows-installer-hash-signature-verification-fallback
• windows-large-vision-request-payload-file
• windows-self-media-publishing-toolbox

draft:

• api-endpoint-connectivity-check
• arkui-scroll-verify-pattern
• codexhost-bridge-stuck-thinking
• codexhost-offline-upgrade
• codexhost-source-merge-upgrade
• codexhost-streaming-bridge
• codexhost-upstream-merge-workflow
• continue-hmharness-conversation-workflow
• harmony-api-symbol-verification
• harmony-arkts-symbol-guard
• harmony-hap-path-resolution
• harmony-launch-install-preflight
• harmony-md-task-executor
• harmony-mock-backend-smoke-test
• harmony-schema-lint-build-loop
• harmony-sdk-symbol-verification
• harmony-staged-task-executor
• harmony-staged-verify-executor
• harmony-tool-failure-env-probe
• harmony-toolchain-preflight
• harmony-verification-task-runner
• harmony-windows-tool-path-resolution
• harmony-windows-toolchain-preflight
• harmony-windows-verify-file-read
• harmony-windows-worktree-tool-path-preflight
• harmony-worktree-path-discovery
• hmh-noninteractive-execution
• hmh-noninteractive-execution-in-codexhost
• hmharness-memory-json-init
• mcp-http-endpoint-config
• read-verify-file-workflow
• scoped-verification-stage-guard
• self-introduction-workflow
• verify-before-execution-workflow
• verify-before-read-workflow
• verify-staged-task-runner
• verify-task-turn-budget-gating
• windows-command-compat-workflow
• windows-curl-download-retry
• windows-harmony-toolchain-path-resolution
• windows-process-port-recovery
• windows-tool-failure-alternative-strategy
• windows-worktree-path-preflight