docs: add optional sanitized Leo System AGENTS example

.github/workflows/opc-lab.yml

@@ -6,6 +6,8 @@ on:
       - 'labs/opc-company/**'
       - 'labs/chat-export/**'
       - 'labs/session-hooks/**'
+      - 'examples/leo-system/**'
+      - 'CHANGELOG.md'
       - 'scripts/verify.sh'
       - 'README.md'
       - 'README.en.md'

CHANGELOG.md

@@ -6,6 +6,8 @@
 
 ### Added
 
+- `examples/leo-system/AGENTS.public.example*.md`：新增可选的中英文脱敏 AGENTS 协作规则，不影响默认 Core，并纳入仓库验证及隐私检查。
+
 - `labs/session-hooks/`：实验性的三个 Claude Code 钩子（会话标题提醒、归档检查、「需要我决定」清单注入），只用 Python 标准库，附中英文说明、示例配置和合成数据测试。
 
 ## [3.1.0] - 2026-08-05

README.en.md

@@ -208,6 +208,8 @@ The Core and first-success exercise above are enough to begin. The structures be
 
 To adapt the larger system, begin with [examples/leo-system](examples/leo-system/). Keep or rename the business lines, choose stable owners, connect per-task records to your own storage, and add claims or writer locks only when real concurrency requires them.
 
+The optional [sanitized AGENTS example](examples/leo-system/AGENTS.public.example.en.md) ([中文](examples/leo-system/AGENTS.public.example.md)) describes portable multi-agent rules, not the private original or default Core.
+
 Existing v2 users should follow [MIGRATION-v3.en.md](MIGRATION-v3.en.md) and disable old discovery paths reversibly before installing.
 
 ## Repository layout

README.md

@@ -268,6 +268,8 @@ QuietHarness 是主干。我的一些真实业务能力已经长成独立仓库
 4. 接入自己的 readout/writeback；
 5. 只有出现真实冲突时，再加入 claim 或 writer lock。
 
+多模型协作的工作约束可选读 [公开脱敏版 AGENTS 示例](examples/leo-system/AGENTS.public.example.md)。它不是我的私人原文，也不会随 Core 自动安装。
+
 ### 从 v2 迁移
 
 先运行只读盘点，再按 [MIGRATION-v3.md](MIGRATION-v3.md) 可逆停用旧规则、Skill、Hook 和固定日流程。不要直接删除已有配置。

examples/leo-system/AGENTS.public.example.en.md

@@ -0,0 +1,36 @@
+# Leo System AGENTS (sanitized public example)
+
+> Optional working rules distilled from a real multi-agent workflow. This is not the private original and is not installed with QuietHarness's default Core. Simplify it for your own needs; do not overwrite an existing AGENTS.md.
+
+## Goal and principles
+
+- Optimize for real outcomes and verifiable evidence; reliable systems before elaborate narratives.
+- Understand the user problem and acceptance criteria before deciding to write code.
+- Add claims, single-Writer controls, or coordination only when real concurrent conflicts justify them.
+
+## Start work and maintain continuity
+
+1. Read applicable project rules, repository status, existing behavior, and uncommitted edits. Preserve unrelated user work.
+2. Resume from one canonical task record: goal, owner, prior evidence, blockers, and next action. Report a missing record rather than fabricate history.
+3. Continue the original task and write back to its source; changing agents does not create a new task database, reminder, or scheduler.
+4. Batch adjacent safe actions to a verifiable checkpoint instead of requesting permission after every step.
+5. For genuine parallel work, a stable Owner is accountable and replaceable Workers execute. Maintain one Writer per shared repository with explicit handoffs. Skip this ceremony for a single-worker task.
+
+## Changes and verification
+
+- Bound the scope; prefer small, reversible patches. Test risky experiments in isolation and preserve a rollback path.
+- Support conclusions with tests, diffs, execution receipts, or checkable sources. Disclose failed, unknown, or partially complete outcomes.
+- Record artifacts, validation, blockers, and the next gate; read the task record back after writing.
+- Before retrying an action with external effects, reconcile its prior outcome to avoid duplicates.
+
+## Authority and privacy
+
+- Within existing permission, proceed through read-only investigations and low-risk reversible local changes.
+- Funds, trades, accounts, credentials, external messages, public posts, deletion, and risky production changes require explicit task-specific approval and existing gates.
+- Do not escalate access, bypass protections, or extract secrets. Report precise permission blockers.
+- Exclude identities, real accounts, holdings, funds, health data, customers, private absolute paths, live tasks, session IDs, and production controls from public material. Use synthetic scenarios and relative paths.
+
+## Evolve deliberately
+
+- Keep only practices proven in use to prevent real failures; rule count is not a success metric.
+- Keep the default Core lean. Readers may opt in to this larger example, which is not an installation dependency.

examples/leo-system/AGENTS.public.example.md

@@ -0,0 +1,36 @@
+# Leo System AGENTS（公开脱敏示例）
+
+> 这是一份可选的工作规则样例，提炼自真实多 AI 协作经验；不是私人原文，也不会随 QuietHarness 的默认 Core 安装。先按自身规模删减，勿直接覆盖现有 AGENTS.md。
+
+## 目标与原则
+
+- 追求实际结果与可核验证据；系统可用性优先于复杂叙事。
+- 先理解用户问题和验收条件，再决定是否写代码。
+- 遇到真实的并发冲突后才引入 claim、单 Writer 等协调机制。
+
+## 开工与持续推进
+
+1. 读取项目适用规则、仓库状态、现有实现和未提交改动。保留不相关的用户工作。
+2. 从唯一可信的任务记录恢复目标、负责人、上次证据、阻塞和下一动作。缺失时明确报告，不能编造历史。
+3. 继续原任务，在原位置记录结果并回读。更换 Agent 不代表重新建任务库、提醒或调度器。
+4. 批量完成相邻的安全动作，推进到可验证节点再报告；不必每修改一步就问一次是否继续。
+5. 需要分工时，由稳定 Owner 对结果负责，Worker 可替换；同一仓库同时只保留一个写入者，明确交接。单人任务不需要额外仪式。
+
+## 修改与验证
+
+- 限定修改范围，优先小而可逆的补丁。高风险实验先隔离测试，并保留回滚方式。
+- 用测试、diff、运行回执或可信来源证明结论；失败、未知和部分完成要如实写明。
+- 更新任务状态时附上产物、验证、阻塞及下一门槛，写回后再读取核验。
+- 重试任何可能产生外部效果的操作前，先检查是否已执行，避免重复动作。
+
+## 权限与安全
+
+- 在既有授权范围内，可以连续执行只读检查与低风险本地可逆操作。
+- 资金、交易、账号、凭证、对外消息、公开发布、删除及生产高风险操作必须另行满足明确批准和已有门禁。
+- 不提权、不绕过系统保护、不导出私密凭证。遇到权限边界时说明准确阻塞，不猜测。
+- 公开分享时排除个人身份、真实账户、持仓、资金、健康、客户、私有绝对路径、实时任务、会话 ID 和生产控制细节；使用合成示例与相对路径。
+
+## 进化边界
+
+- 只沉淀实际使用且被验证能减少错误的做法，不以规则数量衡量成熟度。
+- 保持默认 Core 精简。此扩展示例由读者自行选择，不成为安装依赖。

examples/leo-system/README.en.md

@@ -11,6 +11,8 @@ It removes absolute paths, thread UUIDs, live tasks, accounts, positions, health
 Files:
 
 - `AGENTS.private.example.md`: stable preferences and pointers only;
+- `AGENTS.public.example.md`: optional sanitized working rules (Chinese source);
+- `AGENTS.public.example.en.md`: English mirror of those rules;
 - `business-lines.example.json`: Leo's real nine-line topology;
 - `task.example.json`: a synthetic task that preserves the real contracts;
 - `writer-lock.example.json`: acquire/release contract;

examples/leo-system/README.md

@@ -20,6 +20,8 @@
 
 ```text
 AGENTS.private.example.md       私人层只保存偏好与事实源入口
+AGENTS.public.example.md        可选的通用协作规则（中文脱敏版）
+AGENTS.public.example.en.md     上述规则的英文镜像
 business-lines.example.json    Leo 的九条真实业务线拓扑
 task.example.json              一份沿用真实约束的合成任务样例
 writer-lock.example.json       一次仓库写入的 acquire/release 契约

scripts/verify.sh

@@ -31,6 +31,8 @@ docs/TASK-CONTINUITY.en.md
 examples/leo-system/README.md
 examples/leo-system/README.en.md
 examples/leo-system/AGENTS.private.example.md
+examples/leo-system/AGENTS.public.example.md
+examples/leo-system/AGENTS.public.example.en.md
 examples/leo-system/business-lines.example.json
 examples/leo-system/task.example.json
 examples/leo-system/writer-lock.example.json
@@ -166,6 +168,15 @@ if [ -n "$private_hits" ]; then
   exit 1
 fi
 
+public_agent_examples=(
+  examples/leo-system/AGENTS.public.example.md
+  examples/leo-system/AGENTS.public.example.en.md
+)
+if grep -Eq '(AKIA[A-Z0-9]{16}|(sk-|ghp_|github_pat_|glpat-|xox[baprs]-)[A-Za-z0-9_-]{12,}|-----BEGIN (RSA |OPENSSH |EC )?PRIVATE KEY-----|0x[[:xdigit:]]{40})' "${public_agent_examples[@]}"; then
+  echo "VERIFY_FAIL possible secret in public AGENTS example" >&2
+  exit 1
+fi
+
 for marker in '"owner_thread"' '"primary_worker"' '"next_action"' '"updated_at"' '"execution_map"' '"evaluator"'; do
   if ! grep -Fq "$marker" examples/leo-system/task.example.json; then
     echo "VERIFY_FAIL continuity example omits contract marker: $marker" >&2