LoopX v0.1.16

LoopX v0.1.16

LoopX v0.1.16 is a narrow archive-install provenance hotfix. It does not add a
new product capability.

Fixed

Running loopx update --execute or the no-clone installer from inside an older
LoopX source checkout could produce a mixed result: runtime files came from the
new stable archive, but release.json could import the caller checkout first
and record its older package version. loopx doctor then correctly reported a
manifest/runtime mismatch.

#1776 generates the manifest
from inside the copied release snapshot with an isolated Python path. The
existing no-clone release gate now starts installation from a deliberately stale
checkout and verifies that manifest version, manifest tag, and runtime version
all come from the archive.

Compatibility

• Package version and tag: 0.1.16 / v0.1.16.
• No local-state or manifest-schema migration is required.
• Installer flags, update/rollback semantics, and release layout are unchanged.
• Users who installed v0.1.15 from a normal directory may already have a
  correct manifest. Users who updated while their shell was inside an older
  LoopX checkout should run the upgrade commands below.

Validation

• Main Full Public Smokes passed all 6 shards on the release commit in
  run 29068827711.
• The stale-checkout no-clone regression passed.
• Local installer, packaged installer, fresh-clone quickstart, release version
  contract, update smoke, bash syntax, and Python compile checks passed.
• Standard premerge canary selected 17 checks plus 4 direct checks with
  0 failures, 0 warnings, and no manual holds.
• The changed paths contain no credentials, private state, raw logs, local
  paths, or generated evidence.

Audit basis

The feature/fix range after v0.1.15 contains one code commit, the merge of
#1776. The final release
commit only updates the package version and release timeline.

Full diff: v0.1.15...v0.1.16

中文摘要

v0.1.16 是一个克制的安装来源修复，不新增产品能力。此前如果用户在旧版 LoopX
源码目录中执行 stable 更新，新运行时代码虽然正确，但 release.json 可能误记旧目录的
版本号。本版本隔离 manifest 生成时的工作目录与 Python 导入路径，并把“从旧 checkout
发起安装”纳入 no-clone release gate。无需迁移本地状态。

Upgrade

loopx update --execute
loopx doctor

LoopX v0.1.15

LoopX v0.1.15

Known issue, fixed in v0.1.16:
starting an archive update from inside an older LoopX source checkout could
write that checkout's package version into the new release.json, even
though the installed runtime files were current. Use v0.1.16 as the stable
install/update target.

LoopX v0.1.15 is an actionable-routing and long-run reliability release. It
consolidates the 55 feature, repair, documentation, and validation commits that
landed after v0.1.14, then adds the version/timeline promotion commit. The
stable user-facing theme is clearer agent execution and more dependable
monitor/replan behavior; issue-fix and benchmark/SkillsBench routes remain
experimental.

Highlights

A clearer action contract for agents

• Status and quota now expose the current agent action and selected todo more
  directly, while interaction contracts resolve one primary action through a
  dedicated bounded-context helper (#1720,
  #1731,
  #1751).
• Todo suggestion quality watches, stable reward preview tokens, clearer replan
  spend accounting, and broader recent evidence reads make autonomous replans
  easier to ground and audit (#1732,
  #1733,
  #1736,
  #1737).
• Filtered resumes, due-monitor compaction, replan acknowledgements, terminal
  todo transitions, vision lifecycle states, and auxiliary monitor polling now
  preserve more of the intent needed across bounded turns
  (#1757,
  #1764,
  #1765,
  #1766,
  #1769,
  #1770).

Monitoring and multi-agent continuation

• External dependency observation and Codex CLI task-output quiet timeouts are
  now first-class monitor/worker behavior
  (#1722,
  #1723).
• No-handoff prompts remain in their assigned side-agent lane; host activation
  and heartbeat routes are gated by registered identity and capabilities
  (#1724,
  #1745,
  #1747).
• Review-only closeout no longer depends on domain action-name conventions.
  A small closed continuation_policy contract now distinguishes
  independent_handoff, same_agent_non_delivery, and primary_review, while
  action_kind remains extensible
  (#1754,
  #1773).

Experimental issue-fix workflow

Issue-fix gains PR lifecycle monitoring, a feasibility route projection,
guarded self-repair guidance for proposing GitHub issues, and domain receipts
that distinguish planned from actually completed work
(#1739,
#1742,
#1748,
#1753). These additions help
agents prepare and monitor bounded repair work; they do not grant unattended
production-write or merge authority.

Experimental benchmark and SkillsBench reliability

• Countability and score interpretation are stricter, with a default build
  stall watchdog, compatible curl retries, cache-first bootstrap, proxy runtime
  evidence, and standard aggregate case sets
  (#1726,
  #1734,
  #1738,
  #1741,
  #1744,
  #1746,
  #1756).
• Remote batch closeout now writes standard ledgers with lane-local seeds,
  simpler ownership, omitted-row catch-up, and logically idempotent recovery;
  benchmark projection was also moved into a narrower runtime context
  (#1749,
  #1750,
  #1755,
  #1759,
  #1771,
  #1772).
• Setup/network attribution retains stable fingerprints, while proxy-backed
  prewarming handles base and untagged image destinations more predictably
  (#1760,
  #1761,
  #1762,
  #1763,
  #1767,
  #1768).

These changes improve evidence and recovery plumbing. Benchmark runner behavior,
scoring, upload, and raw task execution remain experimental and are not promoted
as stable public contracts by this release.

Release engineering, install, and documentation

• The repository now has a scheduled/push-triggered full-public smoke workflow,
  a bounded parallel smoke runner, clearer recursive/release-sweep guidance,
  repaired public coverage, and follow-up regression fixes
  (#1718,
  #1721,
  #1727,
  #1728,
  #1729,
  #1740,
  #1743,
  #1752,
  #1774).
• Direct installed invocation is recognized correctly by loopx doctor
  (#1735).
• Explore is presented as experimental and now states why measurable metrics
  are important and why open-ended or weakly quantified work is a poor fit
  (#1725,
  #1730).

Compatibility

• Package version and tag: 0.1.15 / v0.1.15.
• No local-state migration is required.
• Existing extensible action_kind values remain valid. Legacy
  primary_review*, *_review, and *_verification continuation reads remain
  supported, while new writes materialize the typed continuation_policy.
• The default no-clone installer and loopx update continue to use the
  stable ref.

Validation

• Main Full Public Smokes: all 6 shards passed on the release commit in
  run 29068105167.
• Release diff standard premerge canary: 17 selected checks, 4 direct checks,
  0 failures, 0 warnings, and no manual holds.
• Release version/readiness contracts, Python compilation, no-clone release
  verification, fresh-clone quickstart, and update smoke all passed.
• Promotion-readiness canary passed without evidence writeback. Dashboard demo
  readiness was explicitly skipped because optional npm dependencies were not
  installed and this release does not change dashboard/frontstage code.
• Public boundary scan passed across 874 README/docs/examples files with
  0 errors and 0 warnings.
• GitHub reported non-blocking deprecation annotations for action versions that
  declare a Node 20 runtime and are currently forced onto Node 24. No smoke job
  failed because of those annotations.

Audit basis

The feature/fix range v0.1.14..621e7428 contains 55 commits from one author,
touching 182 files with 11,933 insertions and 2,874 deletions. Every commit in
that range is represented in one of the sections above. The final release
commit only updates the package version and release timeline.

Full diff: v0.1.14...v0.1.15

中文摘要

v0.1.15 汇总了 v0.1.14 之后的 55 个功能、修复、文档与验证提交。
稳定能力的重点是：agent 获得更明确的唯一执行动作，monitor/replan/todo 在跨轮次时
更不容易丢失意图，多 agent continuation 由小型枚举策略约束而不再依赖动作字符串命名。
本版本同时补强全量 public smoke、安装诊断和发布门禁。Issue-fix 与
SkillsBench/benchmark 的可行性、证据和恢复链路有较大增量，但仍属于实验能力，
不代表放开无人值守的生产写入、合并、评分或上传权限。

Upgrade

loopx update --check
loopx update --execute
loopx doctor

New installs can continue to use:

curl -fsSL https://raw.githubusercontent.com/huangruiteng/loopx/main/scripts/install-from-github.sh | bash

LoopX v0.1.14

Highlights

LoopX v0.1.14 promotes the developer-contributed software exploration topology work and the latest control-plane reliability fixes to the stable channel.

• Adds the exploration result layer from #1546: public explore node/edge/finding records, graph exports, Lark presentation mapping, and result-log projection for software exploration campaigns.
• Adds opt-in worker/todo branch planning for exploration: the planners are deny-by-default, gated per goal by spawn_policy.explore_harness, and remain read-only unless the normal LoopX quota, todo claim, task lease, writeback, refresh-state, and spend-slot lifecycle authorizes execution.
• Promotes monitor scheduler cadence and quota/status/todo read-model repairs, including quieter monitor backoff, user-gate counts, completed-todo successors, evidence-log counts, delivery lineage, and compact agent-lane status summaries.
• Keeps release distribution on the GitHub stable-channel contract: installer/update flows use the stable ref after promotion; no PyPI publication is required for this release.

Validation

• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 -m py_compile loopx/*.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• python3 examples/explore-worker-plan-gate-smoke.py
• python3 examples/explore-result-layer-smoke.py
• python3 examples/control_plane/quota-plan-smoke.py
• python3 examples/control_plane/status-markdown-smoke.py
• python3 -m loopx.cli check --scan-path README.md --scan-path docs/ --scan-path examples/
• python3 examples/canary/canary-promotion-readiness-smoke.py --no-write-evidence
• python3 -m loopx.cli canary premerge --from-git-diff --tier quick --timeout-seconds 60
• git diff --check

Dashboard browser readiness was not part of this release gate because local dashboard npm dependencies were unavailable; the promotion canary skipped that optional browser subcheck and reported the skip explicitly.

中文摘要

v0.1.14 把开发者贡献的探索拓扑能力提升到 stable：LoopX 现在可以记录探索节点、边和发现，导出图结构，并把结果映射到 Lark 展示面。新的 worker/todo branch planner 默认关闭，只能按 goal 的 spawn_policy.explore_harness 显式开启；即使开启，也仍然必须回到 LoopX 原有的 quota、todo claim、task lease、结果写回和 spend-slot 生命周期里执行。

本版本同时包含 monitor 调度节奏、quota/status/todo 读模型、证据计数、交付 lineage 和 agent-lane 状态摘要等可靠性修复。已有用户可通过 loopx update --check 和 loopx update --execute 更新到该 stable release.

LoopX v0.1.13

Summary

LoopX v0.1.13 promotes the guided onboarding, multi-agent orchestration, scheduler/heartbeat repair, release/update ergonomics, projection modularization, ML task ledger, SkillsBench evidence, and local-control action-policy work merged after v0.1.12.

This is a stable-channel packaging release, not a single-feature drop. It collects the mainline work from PR #1611 through #1647 into the named v0.1.13 version/tag/stable contract.

Diff since v0.1.12

Previous public release: v0.1.12, published 2026-07-08, target commit 5402b472.

Current release: v0.1.13, target commit 6249b6ea.

Diff size:

• 114 files changed
• 8,423 insertions(+)
• 3,271 deletions(-)

Key Changes

• Guided project onboarding and repairability:
  #1631 adds guided start-goal preview,
  #1633 syncs onboarding surfaces,
  #1636 adds non-destructive write-scope migration,
  and #1641 improves refresh-state onboarding diagnostics.
• Multi-agent control plane:
  #1622, contributed by cocolord, routes primary controllers toward subagent orchestration;
  #1643 keeps that feature behind an explicit default-off switch.
• Scheduler and heartbeat reliability:
  #1626 accepts stale monitor scheduler ACK hints,
  #1639 improves scheduler ACK and heartbeat migration,
  #1613 refines monitor scheduler window backoff,
  and #1629 fixes a closed-agent-vision quota gap.
• Projection and rendering maintainability:
  #1640 extracts quota goal-boundary helpers,
  #1642 splits status markdown smoke fixtures,
  #1644 extracts Lark projection row helpers,
  and #1646 extracts content-ops markdown renderers.
• External work and ML experiment support:
  #1627 adds public-safe external ML task ledgers / ML experiment packet support.
• SkillsBench and benchmark-helper hardening:
  #1612 fixes SkillsBench countability and goal launcher behavior,
  #1620 classifies excluded SkillsBench task sources,
  #1621 stabilizes source preflight smoke,
  and #1625 adds the target-lane benchmark ledger aggregate.
• Local-control action text:
  #1645 relaxes next_action / recommended_action validation so local project routing refs are allowed while inline credentials are still rejected.
• Release packaging:
  #1647 bumps the named version to 0.1.13 and adds this stable-channel timeline entry with key PR references.

Benchmark boundary: this release does not publish raw benchmark task text, raw logs, trajectories, verifier output, credentials, leaderboard submission behavior, or scoring changes.

中文摘要

v0.1.13 是一次“接入体验 + 多 agent 控制面 + 调度可靠性 + 投影模块化”的稳定版发布。它不是只发布某一个人的 slice：本版同时包含外部贡献者 cocolord 的 primary-controller subagent orchestration（#1622）、ML experiment / external task ledger（#1627）、guided start/onboarding（#1631、#1633）、write-scope 迁移（#1636）、scheduler/heartbeat 修复（#1626、#1639）、status/quota/Lark/content renderer 拆分（#1640、#1642、#1644、#1646）、SkillsBench evidence hardening（#1612、#1620、#1621、#1625），以及本地控制面 action 文本从 public-safe 硬拦改成只拦 inline credentials（#1645）。

对用户来说，最直接的变化是：新项目接入更顺，scheduler/heartbeat 更不容易因为 ACK 或迁移卡住，主控/子 agent 的分工更可投影，外部实验/任务 evidence 更能进入公开安全的 ledger，状态/投影渲染代码更容易维护，next_action / recommended_action 对本地项目状态更自由但仍保护凭据。

Validation

Release/version gates:

• python3 -m py_compile loopx/*.py
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• python3 examples/install-local-smoke.py
• python3 examples/canary/canary-promotion-readiness-smoke.py --no-write-evidence

Premerge gate:

• loopx canary premerge --from-git-diff --format json --timeout-seconds 120 --no-progress
• result: passed, selected=17, failures=0, manual_holds=0, self_merge_allowed=true

Public/private boundary:

• loopx check --scan-path loopx/__init__.py --scan-path pyproject.toml --scan-path docs/product/release-readiness.md
• changed-file scan: errors=0, warnings=0, public boundary scan clean: 3 files
• promotion-readiness source scan: public boundary scan clean: 1269 files

Known skip:

• Dashboard browser/dependency smoke was skipped by the promotion-readiness smoke because dashboard npm dependencies are not installed in this checkout. The skip is visible and non-blocking for this release boundary; v0.1.13 does not change dashboard code.

Upgrade

For existing stable installs:

loopx update --check
loopx update --dry-run
loopx update --execute
loopx doctor

For repair/fresh install:

curl -fsSL https://raw.githubusercontent.com/huangruiteng/loopx/main/scripts/install-from-github.sh | bash
export PATH="$HOME/.local/bin:$PATH"
loopx doctor

LoopX v0.1.12

Summary

LoopX v0.1.12 promotes the presentation/read-model refactor line, generic frontier/replan recovery fixes, installer overwrite repair, canary progress output, and public benchmark helper contracts to the stable channel.

This is a broad release relative to v0.1.11. The release is not just a version bump: it packages a large body of already-merged control-plane, presentation, install/update, auto-research, and public benchmark-helper work under the named v0.1.12 version/tag/stable contract.

Diff since v0.1.11

Previous public release: v0.1.11, published 2026-07-06, target commit 51491ae4.

Current release: v0.1.12, target commit 5402b472.

Diff size:

• 218 files changed
• 12,309 insertions(+), 6,547 deletions(-)

Main areas:

• Presentation/read-model extraction: large status, goal-channel, dashboard, and Lark rendering paths moved into bounded presentation/read-model modules. This reduces oversized hot files such as loopx/status.py and moves display sinks/renderers into explicit homes.
• Frontier/replan recovery: monitor-only states, open goal-vision gaps, successor/no-follow-up semantics, and quota frontier projection now have more explicit decision-plane coverage and parity smokes.
• Installer repair: local/public installer reruns now overwrite stale wrappers/files safely instead of failing on an existing ~/.local/bin/loopx path.
• Canary iteration speed: premerge canary now streams phase progress earlier, reducing the previous “silent for tens of seconds” feedback gap.
• Auto-research and visible multi-agent flow: visible worker hooks, successor routing, wake/backoff, run-summary reuse, and quiet-completion frontier behavior were tightened.
• Public benchmark helper contracts: route/profile contracts, SkillsBench batch pacing, build-stall timeout, first-action prompt handling, and countable ledger summaries were added or hardened as public helper/control-plane contracts.
• Public docs/community surface: README community QR additions, contributor board refreshes, and release-readiness docs are included from already merged work.

Benchmark boundary: this release does not publish raw benchmark task text, raw logs, trajectories, verifier output, credentials, leaderboard submission behavior, or scoring changes.

Highlights

• More maintainable presentation boundary: status markdown, goal-channel HTML, dashboard routing, and Lark sink rendering have clearer ownership.
• More robust replan behavior: “only monitor todos remain but goal vision is still open” now participates in generic frontier/replan pressure instead of relying on a product-capability-only workaround.
• Safer reinstall/update repair path: users can rerun install/update without manually removing stale loopx wrappers first.
• Better operator feedback during canary: premerge no longer looks stuck while long checks run.
• Cleaner public benchmark helper layer: route/profile and SkillsBench helper contracts are more explicit while keeping raw benchmark evidence out of public release artifacts.

中文摘要

v0.1.12 的主线是把近期已合并的控制面工程质量提升正式发布到 stable：presentation/read-model 边界更清楚，vision/frontier/replan 的通用恢复机制更稳，安装脚本能安全覆盖旧 wrapper，premerge canary 会更早吐进度，auto-research/visible multi-agent 路由更可靠，同时把公开 benchmark helper 合同补齐。

这版不是单点功能发布，而是一次工程质量和可运维性发布。相对 v0.1.11，diff 较大；发布验证因此覆盖了 release/install/update、frontier/replan、presentation read-model、auto-research、benchmark route profile、SkillsBench helper、公共边界扫描和 premerge canary gate。

Validation

Release/version gates:

• python3 -m py_compile loopx/*.py loopx/benchmark_core/*.py
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• python3 examples/install-local-smoke.py
• git diff --check

Focused changed-surface checks:

• python3 -m loopx.cli check --scan-path README.md --scan-path README.zh-CN.md --scan-path docs/ --scan-path examples/
• python3 examples/control_plane/quota-replan-decision-plane-smoke.py
• python3 examples/control_plane/status-collection-readmodel-smoke.py
• python3 examples/control_plane/status-runtime-summaries-readmodel-smoke.py
• python3 examples/control_plane/presentation-markdown-primitives-smoke.py
• python3 examples/canary/premerge-validation-gate-smoke.py
• python3 examples/auto-research-visible-worker-hook-smoke.py
• python3 examples/control_plane/auto-research-role-successor-completion-smoke.py
• python3 examples/benchmark-route-profile-contract-smoke.py
• python3 examples/skillsbench-build-stall-timeout-smoke.py
• python3 examples/skillsbench-batch-case-start-gap-smoke.py

Premerge gate:

• python3 -m loopx.cli canary premerge --from-git-diff
• result: passed, selected=17, failures=0, manual_holds=0, self_merge_allowed=true

Public/private boundary:

• loopx check reported errors=0, warnings=0, public boundary scan clean: 843 files.

Upgrade

For existing stable installs:

loopx update --check
loopx update --dry-run
loopx update --execute
loopx doctor

For repair/fresh install:

curl -fsSL https://raw.githubusercontent.com/huangruiteng/loopx/main/scripts/install-from-github.sh | bash
export PATH="$HOME/.local/bin:$PATH"
loopx doctor

LoopX v0.1.11

Summary

LoopX v0.1.11 promotes the recent vision-replan line to the stable channel.

The main change is that goal vision is no longer just background prose: vision gaps now participate in the quota/replan decision plane, continuation audits are visible to operators, stale vision checkpoint gaps can be superseded by newer closure evidence, and the replan judge has clearer guidance for distinguishing real missing work from stale state.

Highlights since v0.1.10

• Vision-replan decision plane: quota/status can surface vision gaps as actionable replan pressure instead of letting stale checkpoint text strand the loop.
• Continuation audit visibility: vision continuation signals now flow through quota markdown and interaction contracts so operators can see why a lane should continue or stop.
• Stale checkpoint closure: newer evidence can supersede old vision checkpoint gaps, reducing false blockers after a task has already moved on.
• Vision gap judge guidance: the protocol now makes agents decide whether a vision gap is real remaining work, stale state, or a handoff/successor issue.
• Control-plane cleanup: more quota, todo, scheduler, status, public-safety, timestamp, archive, and write-policy logic moved into bounded loopx/control_plane/* contexts with parity smokes.
• Auto-research and agent-state reliability: role-frontier routing, visible evidence checks, successor todo linkage, stale workspace observability, and agent-management status projections were tightened.
• Connector and content ops groundwork: public connector source-map packets and value-connector probes landed for reusable source/evidence mapping.
• CLI/TUI recovery: Codex CLI goal closeout and stale TUI timeout blocker detection were hardened, alongside structured run-index bundle classification.

中文摘要

这版的主线是 vision-replan：goal vision 不再只是文档里的愿景，而是进入 quota/replan 的决策平面。LoopX 现在能把 vision gap 作为可执行的 replan 压力暴露出来，也能在新证据已经关闭问题时 supersede 旧 checkpoint gap，减少“状态已经过期但 agent 还被卡住”的情况。

同时，这版也把近期的控制面重构、auto-research 证据/后继 todo 修复、agent management 只读状态投影、connector source-map、以及 Codex CLI/TUI 恢复路径一起提升到 stable channel。

Validation

• python3 -m py_compile loopx/*.py
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• git diff --check
• python3 -m loopx.cli check --scan-path README.md --scan-path docs/ --scan-path examples/
• python3 examples/canary/canary-promotion-readiness-smoke.py --no-write-evidence
• loopx canary premerge --from-git-diff

Dashboard demo-readiness followed the supported source-checkout skip path because local dashboard npm dependencies were unavailable.

LoopX v0.1.10

Summary

LoopX v0.1.10 promotes the scoped user-gate and agent-management control-plane line to the stable channel.

This release fixes the class of bugs where one controller-owned user gate could accidentally stall unrelated agents. User/owner todos now have an explicit type: user_gate for blocking decisions and user_action for non-blocking owner follow-up. Blocking gates can be scoped to one agent with blocks_agent or declared goal-wide with global_gate=true.

Highlights since v0.1.9

• Scoped owner gates: user todo writes now require explicit user_gate or user_action; quota/status can let independent agent lanes continue when a gate only blocks another agent.
• Agent-management observability: added read-only live agent-management status projections and a dashboard panel surface for inspecting agent lane state without making the UI writable truth.
• Control-plane bounded contexts: continued moving quota, todo, scheduler, review-packet, handoff, and agent identity rules out of broad legacy modules into focused loopx/control_plane/* read models.
• Release/canary hardening: repaired canary fixtures for typed user gates, strengthened premerge validation coverage, and kept promotion readiness tied to public-safe contract checks.
• Codex CLI and benchmark route resilience: improved CLI goal post-bridge recovery, host-local first-action handling, SkillsBench runner source checks, and compact public-safe blocker classification.
• Auto-research reliability: tightened real-run pane closeout, wake handoffs, role-seeded resume chains, and per-agent vision checkpoint behavior.

中文摘要

• 这版主要解决“一个主控 user gate 误伤其他 agent”的问题：用户待办必须明确是阻塞型 user_gate，还是非阻塞型 user_action。
• user_gate 可以通过 blocks_agent 只卡某个 agent，也可以通过 global_gate=true 表示目标级全局 gate；其他不相关 agent lane 可以继续推进。
• 同时把 agent management 的只读状态投影、quota/todo/scheduler/review/handoff 等控制面重构，以及 release/premerge canary 修复一起提升到 stable channel。

Validation

• python3 -m py_compile loopx/*.py
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• python3 -m loopx.cli check --scan-path README.md --scan-path docs/ --scan-path examples/
• python3 examples/canary/canary-promotion-readiness-smoke.py --no-write-evidence
• python3 examples/control_plane/status-markdown-smoke.py
• python3 examples/control_plane/repo-python-line-budget-smoke.py
• loopx canary premerge --from-git-diff

Dashboard demo-readiness ran with browser checks skipped because local dashboard npm dependencies were not installed. The promotion canary treats that as the supported source-checkout --skip-browser path.

LoopX v0.1.9

Summary

LoopX v0.1.9 promotes the current real auto-research and agent-scoped control-plane line to the stable channel.

This release is the first stable cut after the auto-research demo was corrected away from synthetic progress: fake KNN metrics were removed, visible research panes use role-named agent ids, and the KNN preset now runs against a real benchmark workspace with public-safe evidence writeback. It also hardens the replan path so agent lanes can keep moving when completed advancement leaves no successor frontier.

Highlights since v0.1.8

• Real auto-research path: removed fake auto-research demo metrics, slimmed the user/preset entry surfaces, stabilized visible TUI launch, and made reports show actual pane participation rather than precomputed proof.
• KNN research preset: added an explicit knn-demo preset, moved the demo to a real benchmark workspace, normalized evidence, and supports real KNN evidence writeback.
• Agent-scoped evidence: added the agent-scoped evidence ledger protocol, evidence-log CLI/read hints, and replan wiring for required reads.
• Replan/frontier reliability: fixed completed-advancement cases where no successor todo was projected, preserved agent lanes during autonomous replan, and tightened same-title/successor routing.
• Control-plane bounded contexts: moved quota/status/todo/scheduler/runtime/public-safety/projection rules out of legacy broad modules into focused loopx/control_plane/* read models with dedicated smokes.
• Codex CLI and benchmark route hardening: improved CLI goal startup/countability/rate-limit behavior, SkillsBench host-local fixtures, app-goal setup, and post-bridge blocker classification.
• Release and canary safety: added premerge canary gates, release-safe namespace checks, status projection cache canary coverage, and stronger install/update/public-boundary validation.

中文摘要

• 这版把 v0.1.8 之后的一批主线改动收敛到 stable channel，重点是让 LoopX 的长任务、多 agent、安装更新和发布验证路径更稳定、更可追踪。
• Auto-research 相关改动聚焦在真实性：移除预设假指标，保留轻量入口，让 KNN preset 进入真实 benchmark workspace，并把可公开的 evidence 写回控制面。
• Control plane 继续拆分和收紧：quota、status、todo、scheduler、runtime、public-safety 等规则迁到更清晰的 read model，replan/frontier 在无 successor todo 时也更容易恢复。
• Codex CLI、SkillsBench 路由、agent-scoped evidence log 和 canary/premerge gate 都做了加固，目的是减少静默卡住、错误归因和发布后安装/更新漂移。

Update

Existing users can update from the stable channel:

loopx update --check
loopx update --execute
loopx doctor

Fresh install remains:

curl -fsSL https://raw.githubusercontent.com/huangruiteng/loopx/main/scripts/install-from-github.sh | bash
export PATH="$HOME/.local/bin:$PATH"
loopx doctor

Validation

Release-prep PR: #1470

Validated before promotion:

• python3 -m py_compile loopx/*.py
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• PYTHONPATH=. python3 -m loopx.cli check --scan-path README.md --scan-path docs/ --scan-path examples/
• python3 examples/canary/canary-promotion-readiness-smoke.py --no-write-evidence
• PYTHONPATH=. python3 -m loopx.cli canary premerge --from-git-diff

Notes:

• public/private scan was clean for the release diff and public docs/examples scan.
• A local global run-history duplicate-index warning was observed and is not part of this release diff.
• Dashboard canary was explicitly skipped because local npm dashboard dependencies were unavailable; default release canary passed under dashboard auto policy.

LoopX v0.1.8

Summary

LoopX v0.1.8 promotes deterministic agent onboarding and host-loop activation to the stable channel.

The main change is small but important: a new agent host should no longer guess whether codex means Codex App or Codex CLI. Hosts can now ask LoopX for the canonical agent type catalog, pass an exact runtime such as codex-app, codex-cli, or claude-code, and receive the matching host-loop activation contract.

Highlights since v0.1.7

• Added loopx agent-onboard --list-agent-types and loopx agent-onboard --agent-type ... as the deterministic onboarding interface for host runtimes.
• Ambiguous values such as codex now fail closed with concrete suggestions instead of silently falling into the wrong activation path.
• /loopx <task> command packs now explicitly require host-loop activation after todo writeback: Codex App heartbeat automation, Codex CLI visible /goal <task_body>, Claude Code native /loop, or a declared custom host loop.
• Bootstrap command packs now include a host_loop_activation packet and a low-cost recheck policy so normal ticks do not redo onboarding every turn.
• README and protocol docs now distinguish command entry from the underlying loop driver, especially for Codex CLI TUI users.
• Added focused smoke coverage for agent type routing, ambiguous-input rejection, and Codex CLI host-loop activation output.

Update

Existing users can update from the stable channel:

loopx update --check
loopx update --execute
loopx doctor

After updating, refresh host command entries if needed:

loopx slash-commands --install

New hosts can inspect the exact runtime catalog:

loopx agent-onboard --list-agent-types
loopx agent-onboard --agent-type codex-cli --project .

Compatibility notes

• Existing LoopX CLI commands remain compatible.
• Codex App and Codex CLI are intentionally separate agent types because their loop activation mechanisms differ.
• codex is intentionally rejected as an agent type; use codex-app or codex-cli.
• Codex CLI cannot be considered connected by registry/quota identity alone; after todo writeback it must have /goal <task_body> active or show the exact pasteable gate.

Validation

• python3 examples/control_plane/agent-onboard-host-loop-activation-smoke.py
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 -m py_compile loopx/__init__.py loopx/host_loop_activation.py loopx/agent_onboarding.py loopx/bootstrap_command_pack.py loopx/slash_commands.py loopx/slash_command_install.py loopx/cli_commands/starter_bootstrap.py loopx/cli.py examples/control_plane/agent-onboard-host-loop-activation-smoke.py
• python3 -m loopx.cli --format json agent-onboard --agent-type codex --project . returns code 2 with codex-app / codex-cli suggestions.
• python3 -m loopx.cli --format json agent-onboard --agent-type codex-cli --project . --goal-id loopx-meta --task-text 'demo task'
• python3 -m loopx.cli --format json bootstrap-command-pack --project . --host-surface codex-cli-tui --goal-id loopx-meta --goal-text 'demo task'
• python3 -m loopx.cli --format json version reports 0.1.8.
• git diff --check
• python3 -m loopx.cli check --scan-path ... public boundary scans were clean.

loopx check reported the pre-existing loopx-meta duplicate-index warning; it is unrelated to this release. No GitHub Actions checks were reported for the release PRs.

LoopX v0.1.7

Summary

LoopX v0.1.7 promotes the command-entry integration layer to the public stable channel at commit d4f0bf7.

This release is mostly about making the supported "how do I invoke LoopX from my agent host?" path explicit and less magical. Codex does not currently support native custom slash commands, so LoopX now installs explicit Codex command-facade skills such as $loopx instead of pretending /loopx is a native Codex command. Claude Code receives matching skill entries, and the richer LoopX workflow skills stay available for implicit project, PR, doc, social, and self-repair behavior.

Highlights since v0.1.6

• Command-entry install is now a first-class layer: loopx slash-commands --install installs LoopX-managed command facades for Codex and Claude Code.
• Codex command facades are explicit-only (allow_implicit_invocation: false), while workflow skills such as loopx-project, loopx-pr-review, loopx-doc-registry, and loopx-self-repair remain implicitly invokable when the task matches them.
• Legacy LoopX-managed Codex prompt shims are retired during install, while user-owned or unrelated files are left alone.
• README and setup docs now describe the real Codex path: install/connect, then use $loopx <complex task> or /skills; native /loopx remains a future host capability, not a current Codex feature.
• Auto-research and visible-pane startup continued to improve: bootstrap contracts, visible wake readiness, todo index projection, and live pane startup were tightened.
• Control-plane read models were extracted for status, active state metadata, global registry health, usage, event ledger, decision freshness, handoff, attention items, todo summaries, and replan obligations.
• SkillsBench and benchmark operations got safer fail-fast behavior for setup stalls, verifier watchdog coverage, missing-score attribution, and release packaging coverage for the runtime package.

Update

Existing users can update from the stable channel:

loopx update --check
loopx update --execute
loopx doctor

If you want to refresh host command entries after updating, run:

loopx slash-commands --install

New users can install from the stable channel:

curl -fsSL https://raw.githubusercontent.com/huangruiteng/loopx/main/scripts/install-from-github.sh | bash
export PATH="$HOME/.local/bin:$PATH"
loopx doctor

Compatibility notes

• Existing LoopX CLI commands remain compatible.
• Codex users should use $loopx <task> or select the LoopX skill through /skills; Codex native custom slash commands are not available today.
• Claude Code skill entries are installed by the same command-entry installer; deeper Claude hook/adapter installation remains opt-in where applicable.
• The stable ref is expected to point at the v0.1.7 release commit after promotion.

Validation

• python3 -m compileall -q loopx
• python3 examples/release/release-version-contract-smoke.py
• python3 examples/release/release-readiness-doc-smoke.py
• python3 examples/slash-command-install-smoke.py
• python3 examples/slash-command-catalog-smoke.py
• python3 examples/claude-install-optin-smoke.py
• python3 examples/release/codex-cli-no-clone-release-verification-smoke.py
• python3 examples/fresh-clone-quickstart-smoke.py
• python3 examples/loopx-update-smoke.py
• python3 examples/install-local-smoke.py
• python3 examples/cli-help-manpage-smoke.py
• python3 examples/canary/canary-promotion-readiness-smoke.py --no-write-evidence
• git diff --check
• PYTHONPATH=. python3 -m loopx.cli --format json check --scan-path README.md --scan-path docs/ --scan-path examples/ --scan-path loopx/slash_command_install.py --scan-path skills/

loopx check and the promotion canary reported the pre-existing loopx-meta duplicate-index warning; public/private boundary scans were clean. The promotion canary skipped dashboard browser checks because local dashboard npm dependencies were unavailable, matching the existing optional dashboard skip behavior.