Agent Skills › boshu2/agentops › agent-native

agent-native

GitHub

管理代理会话的角色委派与并行执行,定义协调者、实现者等角色职责。提供超时、清理及故障处理机制,强调基于证据的状态观察而非盲目重启,确保隔离性与上下文有效性。

images/gemini/skills/agent-native/SKILL.md boshu2/agentops

Trigger Scenarios

需要并行分发独立任务 配置多角色协作流程 处理代理会话超时或挂起

Install

npx skills add boshu2/agentops --skill agent-native -g -y
More Options

Non-standard path

npx skills add https://github.com/boshu2/agentops/tree/main/images/gemini/skills/agent-native -g -y

Use without installing

npx skills use boshu2/agentops@agent-native

指定 Agent (Claude Code)

npx skills add boshu2/agentops --skill agent-native -a claude-code -g -y

安装 repo 全部 skill

npx skills add boshu2/agentops --all -g -y

预览 repo 内 skill

npx skills add boshu2/agentops --list

SKILL.md

Frontmatter
{
    "name": "agent-native",
    "consumes": [
        "explicit-role-packets"
    ],
    "metadata": {
        "tier": "meta",
        "effects": [
            "manage_runtime_sessions",
            "invoke_selected_executor"
        ],
        "disposition": "keep_optional_adapter",
        "capabilities": [
            "role_dispatch",
            "observe_workers",
            "handoff",
            "dispatch_once"
        ],
        "dependencies": [],
        "canonical_status": "canonical"
    },
    "produces": [
        "runtime-evidence",
        "worker-handoff",
        "per-packet-results"
    ],
    "practices": [
        "team-topologies",
        "design-by-contract"
    ],
    "context_rel": [
        {
            "kind": "customer-of",
            "with": "ntm"
        },
        {
            "kind": "customer-of",
            "with": "codex-exec"
        },
        {
            "kind": "customer-of",
            "with": "agent-mail"
        }
    ],
    "description": "Dispatch independent tasks to parallel workers or selected persistent roles. Use when: delegation is authorized with disjoint scopes; execution does not validate output.",
    "hexagonal_role": "supporting",
    "user-invocable": true,
    "output_contract": "runtime evidence and per-packet candidate, evidence, or error for explicit authorized work",
    "skill_api_version": 1
}

Agent Native

Operate caller-selected agent sessions as explicit roles without turning the runtime into AgentOps lifecycle authority.

For judgment, default to a fresh context in the author's model family. Cross-model Validate, mixed Council and dueling model perspectives are explicit caller selections. Follow references/model-dispatch.md: the working session is the controller; check the explicitly selected adapter at runtime; no factory is required and Agent Mail is never the judgment path. The recipe owns host authorization, finite input/output, timeout and cleanup requirements.

Role requests declare authority; actual native runtime/OS filesystem and egress controls must enforce it. A prompt, worktree, chmod or unrestricted same-user process does not establish isolation. Observe synthetic canary denials before restricted-source work; unavailable protection remains unavailable.

Use native waits or status notifications while workers or checks are pending. Unchanged state is no reason for another analysis, review or provisional retrospective. Observation consumes time and context; it is not free. A known blocking failure deserves action even while other jobs run. For a suspected stall, inspect observable state before choosing a nudge or replacement within authority and remaining bounds. Stop observing at terminal status or the end of the caller's observation window; impatience alone does not justify restart.

Named failure mode — prompt-send optimism: treating a successfully delivered prompt as a working worker; delivery proves transport, not engagement.

For new authorized work after a worker completes, use the selected runtime's documented follow-up or resume operation that starts a turn. A message operation may only queue text for a running worker. Check native state and engagement; do not treat a queued repair request as a resumed implementation attempt.

Anti-pattern: restarting an unresponsive worker as the first move. Corrective: capture its observable state first — a restart destroys the evidence of why it stalled, and rescue is usually cheaper than rerun.

Roles

  • Orchestrator: passes focused intent, scope and evidence references, names the integration/final-review owner, and reports runtime facts. Retrieve extra history only for a consequential uncertainty; a fresh context is not necessarily small. A new goal does not clear history or renew spent bounds.
  • Implementer: may modify only its packet's declared subject.
  • Validator: receives exact candidate content in a fresh, read-only context.
  • Scribe: records runtime evidence without judging acceptance.

Reader and Writer are bounded cheap delegations, not roles with authority: a Reader returns line-referenced bullets over files the caller never loads, and a Writer lands one patterned file from a spec plus a reference file and returns a receipt the caller never reads back. Both are caller-selected per call, default to a cheap model, and yield runtime facts only — a receipt is not validation. For Codex, use the source-owned bulk-reader or code-writer native role (gpt-5.6-luna); pass a fresh bounded task and receive findings or a receipt. The reader uses explicit slices of at most 350 lines; the parent keeps file content out of its context. A reference file is required for a writer. See context-budget delegation for installation, native invocation, opt-in refusal hooks and the limits of role instructions.

Contract

For a caller-selected parallel batch, validate every complete packet before the first launch. Require the selected executor, packet identity and all transitive effects, with canonical workspace-relative write scopes in separate isolation. Resolve symlinks and normalize paths; compare scopes case-insensitively so an alias cannot hide a collision. A lexical disjointness check alone cannot prove symlink or runtime isolation. The reference batch contract rejects nonempty write_scope.exclude because its proof cannot honor those exclusions.

Dispatch each validated packet once and preserve its identity with the result: candidate, evidence or executor error. Do not partly launch a batch that later fails validation, or retry an error as if it had never happened. Native caller authority determines any repair or follow-up. The developer reference scripts/swarm/dispatch_once.py requires an AgentOps source checkout; it is exercised by repository tests and is not bundled with standalone skills. Installed use dispatches through the selected native runtime. This optional batch mode selects no backlog work, creates no queue and integrates no changes.

  1. Require caller intent, role, workspace, authorized source/output scope and evidence destination before starting a worker. Pass source-store/project/work identity and permitted intent locators before execution can fail. Record this dispatch association in caller-owned native comments/metadata or runtime facts, with actual worker session/context IDs explicitly unknown until observed; a requested ID is not an observed ID. This adds no AO packet schema.
  2. Capture observed native runtime/session/context identity at startup, before substantive work and independently of final handoff. Return the observation through the caller-owned native recording channel with its provenance and permitted source locator. Preserve launch failures and unknowns if startup never becomes observable. Follow session associations for separate parent/resume links, supported multi-work spans and frozen source bounds. A controller is not necessarily a native parent; every requested child and resumed execution needs its own observed association. If recording fails, report the gap; do not claim crash recovery from prompt delivery alone. Prove runtime readiness and engagement from observable state; a successful prompt send is not proof of work.
  3. Keep concurrent writers disjoint and isolated. Runtime coordination is not a claim, lease, queue, or completion state in AgentOps.
  4. Record provider state, transcript references, artifacts, and terminal status.
  5. Return runtime evidence to the caller. Do not convert provider retries, reconnects, idle states, or failures into Plan, Candidate, or verdict state.
  6. A validator session may supply judgment to Validate, but only Validate writes verdict.v2. The adapter cannot select AgentOps semantics, issue a binding verdict, or turn factory completion into delivery or validation proof.

NTM, Codex exec, native processes, Agent Mail, and Gas City are replaceable adapters. Use them only when the caller selected that execution shape. A single local agent pays no factory coordination cost. Model identity, when recorded, is a declared runtime fact like context identity — see references/model-dispatch.md.

Native judgment receipts defines exact private receipt references and the independent profile/subject/acceptance checks for caller-required model diversity. Missing native identity never satisfies a leg.

Version History

  • 9f8a711 Current 2026-09-22 10:59

    新增 Codex 原生的 bulk-reader 和 code-writer 角色,支持 gpt-5.6-luna 模型,并添加 PreToolUse Bash 适配器以实施共享读取预算守卫。

  • 8061085 2026-09-09 05:15

    新增原生证据辅助工具与基于作者模型家族的新鲜审查默认值;修复中断Agent的工作会话关联丢失问题。

  • 7b07a7d 2026-08-19 21:58

    对齐 AgentOps 操作层身份,移除旧版飞轮命令及知识状态,重命名核心架构文档,强化一致性检查门控。

  • 3f402e5 2026-07-24 22:06

Same Skill Collection

images/gemini/skills/account-rotation/SKILL.md
images/gemini/skills/agent-mail/SKILL.md
images/gemini/skills/agy-native/SKILL.md
images/gemini/skills/anti-ceremony/SKILL.md
images/gemini/skills/automation-shape-routing/SKILL.md
images/gemini/skills/bootstrap/SKILL.md
images/gemini/skills/cass/SKILL.md
images/gemini/skills/cc-hooks/SKILL.md
images/gemini/skills/codebase-recon/SKILL.md
images/gemini/skills/codex-exec/SKILL.md
images/gemini/skills/converter/SKILL.md
images/gemini/skills/council/SKILL.md
images/gemini/skills/craft-goal/SKILL.md
images/gemini/skills/crank/SKILL.md
images/gemini/skills/dcg/SKILL.md
images/gemini/skills/doc/SKILL.md
images/gemini/skills/domain/SKILL.md
images/gemini/skills/fitness/SKILL.md
images/gemini/skills/goals/SKILL.md
images/gemini/skills/handoff/SKILL.md
images/gemini/skills/human-only-skills/SKILL.md
images/gemini/skills/idea-genie/SKILL.md
images/gemini/skills/implement/SKILL.md
images/gemini/skills/learn/SKILL.md
images/gemini/skills/memory/SKILL.md
images/gemini/skills/ms/SKILL.md
images/gemini/skills/ntm/SKILL.md
images/gemini/skills/one-way-door/SKILL.md
images/gemini/skills/operationalize/SKILL.md
images/gemini/skills/pattern-mining/SKILL.md
images/gemini/skills/plan/SKILL.md
images/gemini/skills/postmortem/SKILL.md
images/gemini/skills/premortem/SKILL.md
images/gemini/skills/product/SKILL.md
images/gemini/skills/rch/SKILL.md
images/gemini/skills/reality-check/SKILL.md
images/gemini/skills/refactor/SKILL.md
images/gemini/skills/research/SKILL.md
images/gemini/skills/reverse-engineer/SKILL.md
images/gemini/skills/route/SKILL.md
images/gemini/skills/rpi/SKILL.md
images/gemini/skills/sbh/SKILL.md
images/gemini/skills/scaffold/SKILL.md
images/gemini/skills/scope/SKILL.md
images/gemini/skills/security/SKILL.md
images/gemini/skills/shared/SKILL.md
images/gemini/skills/skill-builder/SKILL.md
images/gemini/skills/skill-eval/SKILL.md
images/gemini/skills/standards/SKILL.md

Metadata

Files
0
Version
9f8a711
Hash
9b7231d4
Indexed
2026-07-24 22:06

Home - Wiki
Copyright © 2011-2026 iteam. Current version is 2.155.2. UTC+08:00, 2026-09-26 22:59
浙ICP备14020137号-1