Index of every page on this site: llms.txt
All skills
cat-xierluo avatar

/multi-agent-orca

@5fb69a0
Run it yourself

编排两个以上边界独立的本地 worker,使用 Orca Run/Task/Dispatch、独立 worktree/session 或 tmux 回退,由 PM 负责拆解、派发、巡检、429 停滞恢复、独立验收、PR 收口与临时资源清理;也用于用户明确要求“并行推进”“多个 worker”“PM 总控”“Wave Autopilot”或防止 PM 直接实现逃逸。不要用于单个短任务、纯状态同步,或仅需 Git 分支、提交、PR、merge 规则的工作。

Use this Skill: https://skilld.dev/gh/cat-xierluo/legal-skills/multi-agent-orca

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

referenceszcode-gui-monitor-adapter.md

≈2.1k tokens on demand. Your agent reads this file only when SKILL.md points to it.

zcode-gui-monitor-adapter:只读证据 → PM 监控状态适配器

定位

  • 问题:只读会话证据(PR250 采集器 schemaVersion 1 的 metadata JSON)缺少可直接消费的 PM 监控状态和完成待审门。
  • 行为:zcode-gui-monitor-adapter.py 单次读取证据文件 raw bytes(上限 1 MiB)→ 计算 SHA-256 → 严格校验(UTF-8 / JSON / schema / 类型 / 绑定 / 摘要 / 期待 provider+model)→ 固定状态机映射 → 输出固定 JSON。
  • 边界:Python 3.9+ 标准库;零网络、零 daemon、零 Orca 调用、不读 DB、不导入未发布模块。不投递、不监督、不判定真实活性——三个 flags 恒为 false。

用法

# 必填:--evidence / --session-id / --input-id
python3 skills/multi-agent-orca/scripts/zcode-gui-monitor-adapter.py \
  --evidence /path/to/collector-metadata.json \
  --session-id sess_xxx --input-id queue_xxx

# 可选期待值(证据摘要 / 采集环境身份),任一不匹配即固定错误退出
python3 skills/multi-agent-orca/scripts/zcode-gui-monitor-adapter.py \
  --evidence /path/to/collector-metadata.json \
  --session-id sess_xxx --input-id queue_xxx \
  --expected-evidence-sha256 <64位hex> \
  --expected-provider account:bigmodel-start-plan \
  --expected-model GLM-5.3-Flash
参数 必填 说明
--evidence FILE 是 采集器 metadata JSON 路径;只读,绝不创建/写入;>1 MiB 拒绝
--session-id SID 是 必须与 metadata sessionId 完全一致,否则 BINDING_MISMATCH
--input-id INPUT 是 必须与 metadata inputId 完全一致,否则 BINDING_MISMATCH
--expected-evidence-sha256 HEX 否 64 位 hex;与文件 raw bytes 摘要不符 → EVIDENCE_DIGEST_MISMATCH
--expected-provider PROVIDER 否 与 finalAssistant.providerId 不符 → PROVIDER_MISMATCH
--expected-model MODEL 否 与 finalAssistant.modelId 不符 → MODEL_MISMATCH

离线自测:python3 skills/multi-agent-orca/scripts/test-zcode-gui-monitor-adapter.py(真实 CLI 驱动,零网络零副作用;默认不绑定任何远端路径)。真实消费者回归由环境变量显式注入:

ZCODE_GUI_MONITOR_REAL_EVIDENCE=/path/to/real-metadata.json \
ZCODE_GUI_MONITOR_REAL_SESSION=sess_xxx \
ZCODE_GUI_MONITOR_REAL_INPUT=queue_xxx \
# 可选:ZCODE_GUI_MONITOR_REAL_SHA256 / _REAL_PROVIDER / _REAL_MODEL
python3 skills/multi-agent-orca/scripts/test-zcode-gui-monitor-adapter.py

三要素缺失或文件不存在时该块明确打印 SKIP,不影响其余用例。

采集器合法 nullable 约定(schemaVersion 1)

采集器无 turn 时固定输出 turn = {"found": false, "turnId": null, "status": null}。适配器据此把标量字段的 null 视同"缺失/未观测"(input.status、turn.found/status/turnId、finalAssistant.* 布尔与 providerId/modelId、completionEvidence.*),仅接受"非 null 且类型错误"才判 EVIDENCE_TYPE_INVALID。例如 input.status=null + 无 turn → INPUT_ACCEPTED;turn.status=null + found=false → TURN_UNKNOWN。非 unknown 缺值绝不补全为成功。

状态机(固定词表,冲突检测先行)

先做冲突检测(见下节);无冲突后按序映射:

观察 状态
input.status == "admitted"(无 concrete turn/final) INPUT_ACCEPTED
input.status == "failed"(无 concrete turn/final) INPUT_FAILED
input.status 缺失 / null / 未知值(无论 turn/final 观测如何) TURN_UNKNOWN
input promoted 但无 turn(turn.found 非 true) TURN_UNKNOWN
turn.status == "running" / "error" / "cancelled" TURN_RUNNING / TURN_ERROR / TURN_CANCELLED
turn.status 为其他未知值(含 null 且 found=true) TURN_UNKNOWN(保留 UNKNOWN)
turn completed,finalAssistant.errorFree 明确 false DELIVERY_ERROR
turn completed,最终 assistant 或 complete=true 任一不齐 DELIVERY_PENDING
promoted 输入 + turn completed + finalAssistant 四项布尔 true + completionEvidence.complete 显式 true READY_FOR_PM_REVIEW
证据内部矛盾 EVIDENCE_CONFLICT(先行短路)

完成待审门要求已证 promoted 输入:READY_FOR_PM_REVIEW 仅当 input.status 显式为 "promoted" 时可达;input 缺失/未知值即使 turn/final 观测全 true 也只给 TURN_UNKNOWN。READY 另需 completionEvidence.complete 显式 true 且与观测一致——缺项保持未知不补全(落入 DELIVERY_PENDING)。

EVIDENCE_CONFLICT 触发条件(双向一致合同)

存在且可知(非 null)的两侧布尔必须任一方向一致;缺失/null 键保持未知、绝不构成冲突。以下任一成立即 EVIDENCE_CONFLICT(exit 0):

  1. completionEvidence.complete(显式 bool)≠ 观测 complete(turn completed 且 final 四项全 true)——既拒绝伪造 complete=true,也拒绝欠声明 complete=false;
  2. completionEvidence.turnCompleted ≠ 观测 turn completed;
  3. completionEvidence.finalAssistantCompleted / finalAssistantErrorFree / finalTextAvailable(显式 bool)≠ 对应 finalAssistant 布尔为 true 的观测;
  4. finalAssistant.found == false 但 completed / errorFree / textAvailable 任一为 true(未 found 却声明具体布尔);
  5. turn.found == false 但 turn.status 是 running/completed/error/cancelled 之一(turn 自身矛盾);
  6. input.status ∈ {admitted, failed}(已证非 promoted)但 turn.found == true 或 finalAssistant.found == true(concrete turn/final 矛盾)。

正常进行中快照不误拒:running turn + 全 false 的 completionEvidence 一致,照常 TURN_RUNNING;pending 快照(final 不全 + 显式 false 的 ce)照常 DELIVERY_PENDING。旧成功不被掩盖:failed 输入短路,turn error 优先于残留成功 finalAssistant(前者若带 concrete turn 则先行冲突)。

错误契约(固定 JSON + exit 2)

缺文件 / 超 1 MiB / 坏 UTF-8 / 坏 JSON / schema 非 1 / 类型错误 / 绑定不一致 / 摘要不匹配 / 期待 provider 或 model 不一致 → 输出单一固定 JSON 并以 exit 2 退出。错误输出不含路径、不含输入原文、不含 traceback,仅有:

{"adapter": "zcode-gui-monitor-adapter", "schemaVersion": 1, "error": "<错误码>", "flags": {"pmAccepted": false, "orcaSupervised": false, "livenessAuthoritative": false}}

错误码:EVIDENCE_UNREADABLE、EVIDENCE_TOO_LARGE、EVIDENCE_NOT_UTF8、EVIDENCE_NOT_JSON、EVIDENCE_SCHEMA_UNSUPPORTED、EVIDENCE_TYPE_INVALID、EVIDENCE_DIGEST_MISMATCH、BINDING_MISMATCH、PROVIDER_MISMATCH、MODEL_MISMATCH。

输出契约

成功输出(exit 0)只含绑定 IDs、摘要、固定状态与 flags,未知字段一律丢弃:

{
  "adapter": "zcode-gui-monitor-adapter",
  "schemaVersion": 1,
  "sessionId": "sess_...",
  "inputId": "queue_...",
  "turnId": "turn_... | null",
  "evidenceSha256": "<64位hex>",
  "state": "<固定状态>",
  "provider": "<providerId 或 unknown>",
  "model": "<modelId 或 unknown>",
  "flags": {"pmAccepted": false, "orcaSupervised": false, "livenessAuthoritative": false}
}
  • pmAccepted=false:适配器无权代表 PM 接受任何交付;
  • orcaSupervised=false:适配器不参与、不声称任何真实 Orca 监督;
  • livenessAuthoritative=false:适配器不做活性判定,缺失字段绝不推断为 idle/completed;
  • provider/model 缺失时输出 unknown,绝不由期待值回落填充。

cold 快照边界

  • 单次读取、时点快照:证据只反映采集时刻,适配器输出只对该时点负责;同一文件重跑 evidenceSha256 一致,文件内容与目录条目不变。
  • 不推断:metadata 中缺失或 null 的字段一律按"未知"处理并落入 fail-closed 状态(TURN_UNKNOWN/DELIVERY_PENDING),绝不补全为成功。
  • 不可作为投递依据:任何状态(含 READY_FOR_PM_REVIEW)都不能触发自动 GUI 投递或监督动作;真实 Dispatch 必须走原生通道,由 PM 人工复核后另行发起。
  • 不泄露:reasoning、tool 载荷、正文文本、凭证等未消费字段永远不出现在输出中。

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

1 alert18d3 checks · Risk HIGH
  • Gen Agent Trust Hub18d

    This skill is a highly sophisticated multi-agent orchestration framework designed to manage and parallelize local AI worker tasks using tools like Orca, tmux, and SSH. While it incorporates an extensive array of security features—including automated installation guards, file write scope guards, and a structured authority receipt system—it fundamentally relies on broad system permissions. It performs extensive command execution, dynamic script loading, and remote execution on network nodes via SSH. Users should review the configuration and security policies (e.g., config/harness-backend-policy.json) to ensure they align with their environment's security requirements.

  • Socket18d

    2 alerts: gptAnomaly

  • Snyk18d

    Risk: LOW · No issues

Signed by skilld at 5fb69a0. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub yesterday.

Activeupdated last month
Other metadata
metadata
{
  "version": "3.1.1",
  "homepage": "https://github.com/cat-xierluo/legal-skills",
  "author": "杨卫薪律师(微信ywxlaw)"
}

README badge

README badge for cat-xierluo/legal-skills/multi-agent-orca