OpenClaw 基操实践之一—— Agent 团队创建手册:环境检测、结构图、运行机制与验收

cd ~/agent-workspaces/openclaw-testecho "===== E0 target freeze ====="date -Iscat <<'EOF'目标：创建隔离的 OpenClaw 四 Agent 团队。团队：quad-coordinator / quad-planner / quad-executor / quad-reviewer。安全边界：不破坏已验收通过的 triad-agents；不改 Gateway / Feishu / Weixin，直到本地链路验收通过。验收顺序：环境检测 → agent 注册 → direct run → wrapper 本地运行 → hard audit → allowlist → routing → Feishu end-to-end。EOF

# E1：Windows / WSL / Ubuntu 基础环境只读检查# PowerShell 执行：wsl --statuswsl -l -v$PSVersionTable.PSVersionGet-ComputerInfo | Select-Object WindowsProductName, WindowsVersion, OsBuildNumber# WSL Ubuntu Bash 执行：lsb_release -a || cat /etc/os-releaseuname -awhoamipwdhostname -I

ps -p 1 -o pid,comm,argssystemctl --user status --no-pager || truenode -vnpm -vgit --versionjq --versionpython3 --versionwhich openclaw || true

openclaw --versionopenclaw gateway status || trueopenclaw status || trueopenclaw channels status --probe || trueopenclaw config validate || true

openclaw agents list --bindings || trueopenclaw agents bindings || trueopenclaw sessions --all-agents --active 60 --limit 150 || true

openclaw approvals get || trueopenclaw exec-policy show || trueopenclaw config get tools.exec || trueopenclaw config get agents || true

cd ~/agent-workspaces/openclaw-testpwdls -la ~/agent-workspacesfind ~/agent-workspaces -maxdepth 2 -type d | sort | sed -n '1,120p'df -h ~

find ~/agent-workspaces/triad-agents -maxdepth 3 -type f | sort | sed -n '1,120p' || truefind ~/agent-workspaces/triad-lab/runs -maxdepth 1 -type d -name 'triad-v3-*' | sort | tail -n 10 || true

cd ~/agent-workspaces/openclaw-testBACKUP_DIR="$HOME/agent-workspaces/quad-lab/backups/preflight-$(date +%Y%m%d_%H%M%S)"mkdir -p "$BACKUP_DIR"cp -a "$HOME/.openclaw/openclaw.json" "$BACKUP_DIR/openclaw.json.bak" 2>/dev/null || truecp -a "$HOME/.openclaw/exec-approvals.json" "$BACKUP_DIR/exec-approvals.json.bak" 2>/dev/null || trueopenclaw agents list --bindings > "$BACKUP_DIR/agents-list.txt" 2>&1 || trueopenclaw approvals get > "$BACKUP_DIR/approvals.txt" 2>&1 || trueecho "BACKUP_DIR=$BACKUP_DIR"

cd ~/agent-workspaces/openclaw-testecho "===== Q0-0 time marker ====="date -Isecho "===== Q0-1 OpenClaw status ====="openclaw --version || trueopenclaw gateway status || trueopenclaw status || trueecho "===== Q0-2 current agents and bindings ====="openclaw agents list --bindings || trueopenclaw agents bindings || trueecho "===== Q0-3 current approvals / exec policy ====="openclaw approvals get || trueopenclaw exec-policy show || trueecho "===== Q0-4 current sessions/tasks baseline ====="openclaw sessions --all-agents --active 60 --limit 150 || trueopenclaw tasks list --runtime cli --status succeeded || trueopenclaw tasks list --runtime cli --status failed || trueecho "===== Q0-5 config validate ====="openclaw config validate || true

cd ~/agent-workspaces/openclaw-testecho "===== Q1-0 time marker ====="date -IsBASE="$HOME/agent-workspaces/quad-agents"mkdir -p "$BASE"/{coordinator,planner,executor,reviewer}/binmkdir -p "$HOME/agent-workspaces/quad-lab"/{runs,docs,backups}echo "===== Q1-1 create coordinator files ====="cat > "$BASE/coordinator/AGENTS.md" <<'EOF'

你是 quad-coordinator，四 Agent 团队的总调度 Agent。

1. 接收用户任务；2. 判断是否需要四 Agent 工作流；3. 只使用唯一授权 wrapper：quad-dispatch-v1.sh；4. 调度顺序：planner -> executor -> reviewer -> coordinator final；5. 不直接替 planner 规划；6. 不直接替 executor 执行；7. 不绕过 reviewer 宣布完成；8. 不读取凭据；9. 不修改 Gateway、Feishu、Weixin、OpenClaw config；10. 不删除文件。

最终结论必须区分：1. Agent 自我说明；2. wrapper 运行文件；3. OpenClaw backend tasks；4. OpenClaw sessions。硬验收以外部 WSL CLI 的 tasks show / sessions 为准。EOFcat > "$BASE/coordinator/TOOLS.md" <<'EOF'# Quad Coordinator Tools唯一允许的四 Agent 调度入口：~/agent-workspaces/quad-agents/coordinator/bin/quad-dispatch-v1.sh允许：- 调用唯一 wrapper；- 读取 quad-lab/runs 下本轮输出；- 输出 digest final。禁止：- 任意 shell；- 修改 Gateway / Feishu / Weixin / OpenClaw config；- 读取 credentials / token / API key / password / private key；- 删除文件；- 使用未授权脚本。

cat > "$BASE/coordinator/IDENTITY.md" <<'EOF'# IdentityAgent ID: quad-coordinatorTeam: quad-agentsFunction: routing, orchestration, final digest.EOFcat > "$BASE/coordinator/SOUL.md" <<'EOF'# Operating PrincipleCoordinate, do not impersonate.Use evidence, not self-assertion.Prefer digest outputs and external hard audit.EOFcat > "$BASE/coordinator/USER.md" <<'EOF'# User Context

echo "===== Q1-2 create planner files ====="cat > "$BASE/planner/AGENTS.md" <<'EOF'

你是 quad-planner，规划 Agent。

1. 理解原始任务；2. 拆解任务目标；3. 给 executor 定义执行范围；4. 给 reviewer 定义审查标准；5. 标注风险与边界；6. 不执行 shell；7. 不修改文件；8. 不声明任务最终完成。

输出必须包含：1. 任务目标；2. 执行计划；3. executor 输入；4. reviewer 审查标准；5. 风险边界；6. 完成标准。EOFcat > "$BASE/planner/TOOLS.md" <<'EOF'# Quad Planner Tools默认不使用工具。只做任务规划、边界定义、验收标准定义。EOFcat > "$BASE/planner/IDENTITY.md" <<'EOF'# IdentityAgent ID: quad-plannerFunction: task planning and risk boundary.EOFcat > "$BASE/planner/SOUL.md" <<'EOF'# Operating PrinciplePlan before execution.Define boundaries before action.EOFecho "===== Q1-3 create executor files ====="cat > "$BASE/executor/AGENTS.md" <<'EOF'

你是 quad-executor，执行 Agent。

1. 只根据 planner 计划执行；2. 输出执行侧分析或受控结果；3. 不读取凭据；4. 不修改 Gateway、Feishu、Weixin、OpenClaw config；5. 不删除文件；6. 不越过 reviewer 宣布完成；7. 不判断整个四 Agent 链路是否成立。

输出必须包含：1. 接收到的任务；2. 执行步骤；3. 执行结果；4. 风险或未完成项；5. 是否遵守边界。EOFcat > "$BASE/executor/TOOLS.md" <<'EOF'# Quad Executor Tools默认不使用 shell。如未来允许执行，必须由 coordinator wrapper 和 allowlist 控制。EOFcat > "$BASE/executor/IDENTITY.md" <<'EOF'# IdentityAgent ID: quad-executorFunction: controlled execution.EOFcat > "$BASE/executor/SOUL.md" <<'EOF'# Operating PrincipleExecute only within assigned scope.Report uncertainty.Do not overclaim.EOFecho "===== Q1-4 create reviewer files ====="cat > "$BASE/reviewer/AGENTS.md" <<'EOF'

你是 quad-reviewer，审查 Agent。

1. 审查 executor 是否遵守 planner 计划；2. 审查是否越权；3. 审查是否遗漏关键风险；4. 审查输出是否足够完整；5. 给出通过、有条件通过或不通过；6. 不替 executor 执行；7. 不修改文件；8. 不读取凭据。

输出必须包含：1. 审查对象；2. 审查标准；3. 发现的问题；4. 风险等级；5. 通过/有条件通过/不通过；6. 给 coordinator 的建议。EOFcat > "$BASE/reviewer/TOOLS.md" <<'EOF'# Quad Reviewer Tools默认不使用 shell。只审查 planner/executor 输出与边界。EOFcat > "$BASE/reviewer/IDENTITY.md" <<'EOF'# IdentityAgent ID: quad-reviewerFunction: audit and risk review.EOFcat > "$BASE/reviewer/SOUL.md" <<'EOF'# Operating PrincipleReview evidence, not intention.Do not approve vague results.EOFecho "===== Q1-5 init git baselines ====="for d in coordinator planner executor reviewerdocd "$BASE/$d"git init >/dev/null 2>&1 || truegit add .git commit -m "init quad $d agent rules" >/dev/null 2>&1 || truedoneecho "===== Q1-6 inspect files ====="find "$BASE" -maxdepth 2 -type f | sort

cd ~/agent-workspaces/openclaw-testecho "===== Q2-0 time marker ====="date -IsBASE="$HOME/agent-workspaces/quad-agents"MODEL="deepseek/deepseek-v4-flash"echo "===== Q2-1 backup OpenClaw config before agents add ====="BACKUP_DIR="$HOME/agent-workspaces/quad-lab/backups/q2-agents-add-$(date +%Y%m%d_%H%M%S)"mkdir -p "$BACKUP_DIR"cp -a "$HOME/.openclaw/openclaw.json" "$BACKUP_DIR/openclaw.json.bak" 2>/dev/null || trueecho "BACKUP_DIR=$BACKUP_DIR"echo "===== Q2-2 add quad agents ====="openclaw agents add quad-coordinator \--workspace "$BASE/coordinator" \--agent-dir "$HOME/.openclaw/agents/quad-coordinator/agent" \--model "$MODEL" \--non-interactive || trueopenclaw agents add quad-planner \--workspace "$BASE/planner" \--agent-dir "$HOME/.openclaw/agents/quad-planner/agent" \--model "$MODEL" \--non-interactive || trueopenclaw agents add quad-executor \--workspace "$BASE/executor" \--agent-dir "$HOME/.openclaw/agents/quad-executor/agent" \--model "$MODEL" \--non-interactive || trueopenclaw agents add quad-reviewer \--workspace "$BASE/reviewer" \--agent-dir "$HOME/.openclaw/agents/quad-reviewer/agent" \--model "$MODEL" \--non-interactive || trueecho "===== Q2-3 list agents ====="openclaw agents list --bindings || trueecho "===== Q2-4 config validate ====="openclaw config validate || true

cd ~/agent-workspaces/openclaw-testecho "===== Q3-0 time marker ====="date -IsRUN_TAG="quad-direct-$(date +%Y%m%d_%H%M%S)"echo "RUN_TAG=$RUN_TAG"openclaw agent --agent quad-coordinator --session-key "agent:quad-coordinator:${RUN_TAG}-coordinator" -m "【Q3 direct test】请说明你是 quad-coordinator。只输出身份、职责、禁止事项。不执行 shell，不修改文件。"openclaw agent --agent quad-planner --session-key "agent:quad-planner:${RUN_TAG}-planner" -m "【Q3 direct test】请说明你是 quad-planner。为一个虚拟任务给出三步计划。不执行 shell，不修改文件。"openclaw agent --agent quad-executor --session-key "agent:quad-executor:${RUN_TAG}-executor" -m "【Q3 direct test】请说明你是 quad-executor。列出收到的任务，不执行 shell，不修改文件。"openclaw agent --agent quad-reviewer --session-key "agent:quad-reviewer:${RUN_TAG}-reviewer" -m "【Q3 direct test】请说明你是 quad-reviewer。审查虚拟 executor 输出：未执行 shell、未修改文件、仅输出说明。给出通过/不通过。"openclaw sessions --agent quad-coordinator --active 30 --limit 50 || trueopenclaw sessions --agent quad-planner --active 30 --limit 50 || trueopenclaw sessions --agent quad-executor --active 30 --limit 50 || trueopenclaw sessions --agent quad-reviewer --active 30 --limit 50 || trueopenclaw sessions --all-agents --active 30 --limit 150 || trueopenclaw tasks show "agent:quad-coordinator:${RUN_TAG}-coordinator" || trueopenclaw tasks show "agent:quad-planner:${RUN_TAG}-planner" || trueopenclaw tasks show "agent:quad-executor:${RUN_TAG}-executor" || trueopenclaw tasks show "agent:quad-reviewer:${RUN_TAG}-reviewer" || true

cd ~/agent-workspaces/openclaw-testecho "===== Q4-0 time marker ====="date -IsWRAPPER="$HOME/agent-workspaces/quad-agents/coordinator/bin/quad-dispatch-v1.sh"mkdir -p "$(dirname "$WRAPPER")"cat > "$WRAPPER" <<'EOF'#!/usr/bin/env bashset -euo pipefailRUN_ID="quad-v1-$(date +%Y%m%d_%H%M%S)"BASE_DIR="$HOME/agent-workspaces/quad-lab/runs/$RUN_ID"mkdir -p "$BASE_DIR"TASK_TEXT="${*:-}"if [ -z "$TASK_TEXT" ]; thenTASK_TEXT="四 Agent wrapper 默认测试：planner 规划，executor 执行，reviewer 审查，coordinator final 汇总。"fiTASK_FILE="$BASE_DIR/task.txt"META_FILE="$BASE_DIR/meta.txt"PLANNER_RAW="$BASE_DIR/planner.raw.md"PLANNER_CLEAN="$BASE_DIR/planner.clean.md"PLANNER_DIGEST="$BASE_DIR/planner.digest.md"EXECUTOR_RAW="$BASE_DIR/executor.raw.md"EXECUTOR_CLEAN="$BASE_DIR/executor.clean.md"EXECUTOR_DIGEST="$BASE_DIR/executor.digest.md"REVIEWER_RAW="$BASE_DIR/reviewer.raw.md"REVIEWER_CLEAN="$BASE_DIR/reviewer.clean.md"REVIEWER_DIGEST="$BASE_DIR/reviewer.digest.md"FINAL_PROMPT="$BASE_DIR/coordinator.prompt.txt"FINAL_RAW="$BASE_DIR/coordinator.raw.md"FINAL_CLEAN="$BASE_DIR/coordinator.final.md"PLANNER_SESSION="agent:quad-planner:${RUN_ID}-planner"EXECUTOR_SESSION="agent:quad-executor:${RUN_ID}-executor"REVIEWER_SESSION="agent:quad-reviewer:${RUN_ID}-reviewer"COORDINATOR_SESSION="agent:quad-coordinator:${RUN_ID}-final"clean_openclaw_output() {local input="$1"local output="$2"awk '/Doctor warnings/ {next}/state-migrations/ {next}/Left plugin install index/ {next}/shared SQLite state/ {next}/conflicting plugin install metadata/ {next}/openclaw-weixin/ {next}/^OpenClaw [0-9]/ {next}/^[[:space:]]*│/ {next}/^[[:space:]]*◇/ {next}/^[[:space:]]*├/ {next}/^[[:space:]]*╰/ {next}/^[[:space:]]*╮/ {next}/^[[:space:]]*─/ {next}{print}' "$input" > "$output"}digest_for_prompt() {local input="$1"local head_bytes="${2:-1600}"local tail_bytes="${3:-700}"if [ ! -f "$input" ]; thenecho "[MISSING_FILE] $input"return 0filocal bytesbytes="$(wc -c < "$input" | tr -d ' ')"echo "[FILE] $input"echo "[BYTES] $bytes"echo "[DIGEST_MODE] head_bytes=$head_bytes tail_bytes=$tail_bytes"if [ "$bytes" -le $((head_bytes + tail_bytes + 300)) ]; thencat "$input"elseecho "[HEAD]"head -c "$head_bytes" "$input"printf "\n[...OMITTED_FOR_DIGEST...]\n"echo "[TAIL]"tail -c "$tail_bytes" "$input"printf "\n[DIGEST_END]\n"fi}{echo "RUN_ID=$RUN_ID"echo "BASE_DIR=$BASE_DIR"echo "PLANNER_SESSION=$PLANNER_SESSION"echo "EXECUTOR_SESSION=$EXECUTOR_SESSION"echo "REVIEWER_SESSION=$REVIEWER_SESSION"echo "COORDINATOR_SESSION=$COORDINATOR_SESSION"echo "CREATED_AT=$(date -Is)"echo "WRAPPER=quad-dispatch-v1.sh"echo "CLEANING=enabled"echo "DIGEST=enabled"} > "$META_FILE"printf "%s\n" "$TASK_TEXT" > "$TASK_FILE"echo "===== QUAD V1 RUN $RUN_ID / planner ====="openclaw agent --agent quad-planner --session-key "$PLANNER_SESSION" -m "

你是 quad-planner，规划 Agent。原始任务：$TASK_TEXT要求：1. 不执行 shell；2. 不修改文件；3. 输出任务拆解、executor 输入、reviewer 审查标准、风险边界、完成标准；4. 不声明最终完成。" 2>&1 | tee "$PLANNER_RAW"clean_openclaw_output "$PLANNER_RAW" "$PLANNER_CLEAN"digest_for_prompt "$PLANNER_CLEAN" 1800 800 > "$PLANNER_DIGEST"echoecho "===== QUAD V1 RUN $RUN_ID / executor ====="openclaw agent --agent quad-executor --session-key "$EXECUTOR_SESSION" -m "

你是 quad-executor，执行 Agent。原始任务：$TASK_TEXTplanner digest：$(cat "$PLANNER_DIGEST")要求：1. 只根据 planner digest 执行；2. 不执行 shell；3. 不修改文件；4. 输出接收到的任务、执行步骤、执行结果、风险或未完成项、是否遵守边界；5. 不判断整条链路是否成立。" 2>&1 | tee "$EXECUTOR_RAW"clean_openclaw_output "$EXECUTOR_RAW" "$EXECUTOR_CLEAN"digest_for_prompt "$EXECUTOR_CLEAN" 1800 800 > "$EXECUTOR_DIGEST"echoecho "===== QUAD V1 RUN $RUN_ID / reviewer ====="openclaw agent --agent quad-reviewer --session-key "$REVIEWER_SESSION" -m "

你是 quad-reviewer，审查 Agent。原始任务：$TASK_TEXTplanner digest：$(cat "$PLANNER_DIGEST")executor digest：$(cat "$EXECUTOR_DIGEST")要求：1. 审查 executor 是否遵守 planner；2. 审查是否越权；3. 审查是否遗漏关键风险；4. 输出审查对象、审查标准、发现问题、风险等级、通过/有条件通过/不通过；5. 不执行 shell，不修改文件。" 2>&1 | tee "$REVIEWER_RAW"clean_openclaw_output "$REVIEWER_RAW" "$REVIEWER_CLEAN"digest_for_prompt "$REVIEWER_CLEAN" 1800 800 > "$REVIEWER_DIGEST"echoecho "===== QUAD V1 RUN $RUN_ID / coordinator final ====="cat > "$FINAL_PROMPT" <        <EOF_FINAL< span>         </EOF_FINAL<>

你是 quad-coordinator，总调度 Agent。本轮采用 digest 模式：- 不嵌入 planner/executor/reviewer clean 全文；- 只读取 digest；- 最终硬验收以外部 WSL CLI 的 openclaw tasks show / sessions 为准。外部调度事实：- RUN_ID: $RUN_ID- planner session: $PLANNER_SESSION- executor session: $EXECUTOR_SESSION- reviewer session: $REVIEWER_SESSION- coordinator session: $COORDINATOR_SESSION- run dir: $BASE_DIR原始任务 digest：$(printf "%s" "$TASK_TEXT" | head -c 800)planner digest：$(cat "$PLANNER_DIGEST")executor digest：$(cat "$EXECUTOR_DIGEST")reviewer digest：$(cat "$REVIEWER_DIGEST")输出要求：1. 当前任务目标，3 行以内；2. planner 规划结论；3. executor 执行结论；4. reviewer 审查结论；5. 最多 6 条可执行清单；6. 明确外部硬验收命令；7. 不得声称 backend task 或 direct session 不适用。EOF_FINALopenclaw agent --agent quad-coordinator --session-key "$COORDINATOR_SESSION" -m "$(cat "$FINAL_PROMPT")" 2>&1 | tee "$FINAL_RAW"clean_openclaw_output "$FINAL_RAW" "$FINAL_CLEAN"echoecho "===== QUAD V1 RUN COMPLETED ====="echo "RUN_ID=$RUN_ID"echo "RUN_DIR=$BASE_DIR"echo "META_FILE=$META_FILE"echo "FINAL_CLEAN=$FINAL_CLEAN"EOFchmod +x "$WRAPPER"ls -l "$WRAPPER"grep -n "QUAD V1 RUN\|digest_for_prompt\|quad-planner\|quad-executor\|quad-reviewer\|quad-coordinator" "$WRAPPER" || true

cd ~/agent-workspaces/openclaw-testecho "===== Q5-0 local quad wrapper test ====="date -Is~/agent-workspaces/quad-agents/coordinator/bin/quad-dispatch-v1.sh \"

cd ~/agent-workspaces/openclaw-testecho "===== Q6-0 local quad verification ====="date -IsLATEST_RUN="$(find ~/agent-workspaces/quad-lab/runs -maxdepth 1 -type d -name 'quad-v1-*' | sort | tail -n 1)"RUN_ID="$(basename "$LATEST_RUN")"echo "LATEST_RUN=$LATEST_RUN"echo "RUN_ID=$RUN_ID"find "$LATEST_RUN" -maxdepth 1 -type f -printf '%TY-%Tm-%Td %TH:%TM %p\n' | sortcat "$LATEST_RUN/meta.txt"openclaw tasks show "agent:quad-planner:${RUN_ID}-planner" || trueopenclaw tasks show "agent:quad-executor:${RUN_ID}-executor" || trueopenclaw tasks show "agent:quad-reviewer:${RUN_ID}-reviewer" || trueopenclaw tasks show "agent:quad-coordinator:${RUN_ID}-final" || trueopenclaw sessions --agent quad-planner --active 60 --limit 80 || trueopenclaw sessions --agent quad-executor --active 60 --limit 80 || trueopenclaw sessions --agent quad-reviewer --active 60 --limit 80 || trueopenclaw sessions --agent quad-coordinator --active 60 --limit 80 || trueopenclaw sessions --all-agents --active 60 --limit 200 || truegrep -RInE "Doctor warnings|state-migrations|Left plugin install index|shared SQLite state|conflicting plugin install metadata|openclaw-weixin" \"$LATEST_RUN"/*.clean.md "$LATEST_RUN"/coordinator.final.md 2>/dev/null \|| echo "CLEAN_OK: no OpenClaw warning text found in clean outputs"grep -RInE "backend task.*不适用|direct session.*不适用|not applicable|不创建后端 task|不产生 backend sessions" \"$LATEST_RUN"/coordinator.final.md 2>/dev/null \|| echo "WORDING_OK: no forbidden backend/session misjudgment found"sed -n '1,220p' "$LATEST_RUN/coordinator.final.md" 2>/dev/null || echo "MISSING coordinator.final.md"

cd ~/agent-workspaces/openclaw-testecho "===== Q7-0 time marker ====="date -IsWRAPPER="$HOME/agent-workspaces/quad-agents/coordinator/bin/quad-dispatch-v1.sh"APPROVALS_FILE="$HOME/.openclaw/exec-approvals.json"BACKUP_DIR="$HOME/agent-workspaces/quad-lab/backups/q7-approvals-$(date +%Y%m%d_%H%M%S)"mkdir -p "$BACKUP_DIR"if [ -f "$APPROVALS_FILE" ]; thencp -a "$APPROVALS_FILE" "$BACKUP_DIR/exec-approvals.json.bak"echo "BACKUP_OK $APPROVALS_FILE"elseecho "MISSING $APPROVALS_FILE"fiopenclaw approvals allowlist add --agent quad-coordinator "$WRAPPER" || trueopenclaw approvals get || truegrep -RIn "quad-dispatch-v1.sh\|quad-coordinator" "$APPROVALS_FILE" 2>/dev/null || true

cd ~/agent-workspaces/openclaw-testecho "===== Q8-0 Feishu route precheck ====="date -Isopenclaw agents bindings || trueopenclaw agents list --bindings || trueopenclaw agents bind --help || trueopenclaw agents unbind --help || trueopenclaw channels status --probe || true

cd ~/agent-workspaces/openclaw-testecho "===== Q9-0 backup before route switch ====="BACKUP_DIR="$HOME/agent-workspaces/quad-lab/backups/q9-route-switch-$(date +%Y%m%d_%H%M%S)"mkdir -p "$BACKUP_DIR"cp -a "$HOME/.openclaw/openclaw.json" "$BACKUP_DIR/openclaw.json.bak" 2>/dev/null || true# 具体 unbind 参数以本机 `openclaw agents unbind --help` 为准。openclaw agents unbind --agent coordinator --bind feishu:default || trueopenclaw agents bind --agent quad-coordinator --bind feishu:default || trueopenclaw agents bindings || trueopenclaw agents list --bindings || trueopenclaw channels status --probe || true

请作为 quad-coordinator 处理本任务。请使用唯一授权 wrapper：~/agent-workspaces/quad-agents/coordinator/bin/quad-dispatch-v1.sh传入任务文本：【Q10 四 Agent 真实小任务】请让 planner 规划当前四 Agent 工作流的维护任务；executor 提出 3 个维护风险；reviewer 审查 executor 是否越权、是否遗漏关键风险；最后 coordinator final 使用 digest 模式输出最多 6 条可执行维护清单。要求不读取凭据、不修改配置、不删除文件。约束：1. 只能使用 quad-dispatch-v1.sh。2. 不得执行其他 shell。3. 不得修改 Gateway、Feishu binding、Weixin binding、OpenClaw config。4. 不得读取 credentials、token、API key、password、private key。5. 不得输出长篇全文，只输出 digest 结论。6. 不得声称 backend task 或 direct session 不适用；如果本地视角看不到其他 agent session，只能说明需要外部 WSL 审计。

cd ~/agent-workspaces/openclaw-testecho "===== Q10-1 after Feishu quad regression ====="date -IsLATEST_RUN="$(find ~/agent-workspaces/quad-lab/runs -maxdepth 1 -type d -name 'quad-v1-*' | sort | tail -n 1)"RUN_ID="$(basename "$LATEST_RUN")"DOC_DIR="$HOME/agent-workspaces/quad-lab/docs"AUDIT_FILE="$DOC_DIR/Q10_final_feishu_quad_audit_$(date +%Y%m%d_%H%M%S).md"mkdir -p "$DOC_DIR"echo "LATEST_RUN=$LATEST_RUN"echo "RUN_ID=$RUN_ID"echo "AUDIT_FILE=$AUDIT_FILE"{echo "# Q10 Final Feishu Quad-Agent Audit"echoecho "Generated at: $(date -Is)"echoecho "## Run"echo "- RUN_ID: $RUN_ID"echo "- RUN_DIR: $LATEST_RUN"echoecho "## Files"find "$LATEST_RUN" -maxdepth 1 -type f -printf '- %f\n' | sortechoecho "## Meta"cat "$LATEST_RUN/meta.txt" 2>/dev/null || echo "MISSING meta.txt"echoecho "## Task status"echo "### planner"openclaw tasks show "agent:quad-planner:${RUN_ID}-planner" || trueechoecho "### executor"openclaw tasks show "agent:quad-executor:${RUN_ID}-executor" || trueechoecho "### reviewer"openclaw tasks show "agent:quad-reviewer:${RUN_ID}-reviewer" || trueechoecho "### coordinator"openclaw tasks show "agent:quad-coordinator:${RUN_ID}-final" || trueechoecho "## Sessions"openclaw sessions --agent quad-planner --active 60 --limit 80 || trueopenclaw sessions --agent quad-executor --active 60 --limit 80 || trueopenclaw sessions --agent quad-reviewer --active 60 --limit 80 || trueopenclaw sessions --agent quad-coordinator --active 60 --limit 80 || trueechoecho "## Clean warning check"grep -RInE "Doctor warnings|state-migrations|Left plugin install index|shared SQLite state|conflicting plugin install metadata|openclaw-weixin" \"$LATEST_RUN"/*.clean.md "$LATEST_RUN"/coordinator.final.md 2>/dev/null \|| echo "CLEAN_OK"echoecho "## Forbidden wording check"grep -RInE "backend task.*不适用|direct session.*不适用|not applicable|不创建后端 task|不产生 backend sessions" \"$LATEST_RUN"/coordinator.final.md 2>/dev/null \|| echo "WORDING_OK"echoecho "## Final digest"sed -n '1,220p' "$LATEST_RUN/coordinator.final.md" 2>/dev/null || echo "MISSING coordinator.final.md"} | tee "$AUDIT_FILE"echo "===== Q10-2 screen digest ====="echo "RUN_ID=$RUN_ID"echo "AUDIT_FILE=$AUDIT_FILE"echo "--- required files ---"for f in task.txt meta.txt planner.clean.md planner.digest.md executor.clean.md executor.digest.md reviewer.clean.md reviewer.digest.md coordinator.prompt.txt coordinator.raw.md coordinator.final.mddotest -f "$LATEST_RUN/$f" && echo "OK $f" || echo "MISSING $f"doneecho "--- final pass/fail hints ---"grep -q "status: succeeded" "$AUDIT_FILE" && echo "TASK_EVIDENCE_PRESENT" || echo "TASK_EVIDENCE_NEEDS_REVIEW"grep -q "CLEAN_OK" "$AUDIT_FILE" && echo "CLEAN_OK" || echo "CLEAN_CHECK_FAILED"grep -q "WORDING_OK" "$AUDIT_FILE" && echo "WORDING_OK" || echo "WORDING_CHECK_REVIEW"test -f "$LATEST_RUN/coordinator.final.md" && echo "FINAL_FILE_OK" || echo "FINAL_FILE_MISSING"

task.txtmeta.txtplanner.raw.md / planner.clean.md / planner.digest.mdexecutor.raw.md / executor.clean.md / executor.digest.mdreviewer.raw.md / reviewer.clean.md / reviewer.digest.mdcoordinator.prompt.txtcoordinator.raw.mdcoordinator.final.md