菜单
Create, list, inspect, remove, and garbage-collect isolated VAWS agent sessions. Use before remote execution when multiple agent tasks must run in parallel without sharing local worktrees, remote containers, serving state, or resource leases.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | session-management |
| description | Create, list, inspect, remove, and garbage-collect isolated VAWS agent sessions. Use before remote execution when multiple agent tasks must run in parallel without sharing local worktrees, remote containers, serving state, or resource leases. |
Create and maintain isolated VAWS sessions for parallel agent work.
Each session binds:
.vaws-local/sessions/<session-id>/ state namespace--session-id, VAWS_SESSION_ID, or VAWS_AGENT_SESSION_ID when an upstream agent already has a stable id.session_create.py creates a fresh generated id when no explicit/env id is provided; it does not reuse .vaws-local/current-session.json as a creation default..vaws-local/current-session.json as a convenience fallback.session_create.py.--devices or --npu-count; session-aware serving uses that lease by default.--session-id <id> or --session-file <session.json> to parity, serving, benchmark, profiling-collection, memory-profiling, and profiling-analysis entry points.serve_stop.py --machine <alias> from a session-scoped task.python3 .agents/skills/session-management/scripts/session_create.py \
--machine <alias-or-ip> \
[--session-id <id>] \
[--base-ref main] \
[--branch session/<id>] \
[--devices 0,1] \
[--npu-count 2] \
[--verification-mode ssh|full] \
[--disable-prepared-image-cache]
python3 .agents/skills/session-management/scripts/session_list.py
python3 .agents/skills/session-management/scripts/session_status.py --session-id <id>
python3 .agents/skills/session-management/scripts/session_remove.py --session-id <id> --remove-container --remove-worktree --release-leases
python3 .agents/skills/session-management/scripts/session_gc.py
Progress is emitted on stderr as __VAWS_SESSION_PROGRESS__=<json>. Final output is JSON on stdout.
By default session creation uses a host-local prepared image cache keyed by the selected base image id. The first session for a base image may still install container SSH packages, then commits vaws-session-prepared:<image-hash>-ssh-v2; later sessions start from that prepared image and skip the repeated openssh package install and cached pip/pytest bootstrap. Use --disable-prepared-image-cache only when validating raw base-image bootstrap behavior.
Session creation defaults to --verification-mode ssh: it verifies host SSH and direct session-container SSH, then leaves NPU runtime proof to the task that actually uses the session, such as serving, benchmark, or profiling. Use --verification-mode full when validating a raw machine/container bootstrap and you need the extra torch / torch_npu smoke check during creation.
Local untracked state lives under .vaws-local/sessions/:
index.jsonleases.jsonlocks/<session-id>/session.json<session-id>/serving.json<session-id>/benchmark/Worktree bindings are written to <worktree>/.vaws-local/current-session.json and include the absolute base session file path so scripts run from the worktree can find the base session state.
For explicit --session-id --no-worktree timing/debug sessions, session_create.py does not overwrite the repo-root .vaws-local/current-session.json; agents should pass --session-id or --session-file explicitly for those shared-root flows. Current-session binding writes are atomic so readers never observe partial JSON.
.agents/skills/session-management/references/behavior.md.agents/skills/session-management/references/command-recipes.md.agents/skills/session-management/references/acceptance.mdAdd, verify, repair, or remove a managed remote NPU host for this workspace. Use for requests like “配置服务器”, “加一台机器”, “检查 ready”, “修容器 SSH”, or “移除机器”. Do not use for code sync, rebuilds, serving, or benchmarking.
Collect one Ascend torch-profiler case end-to-end on a workspace-managed remote NPU container. Starts a profiled vLLM service, brackets a workload with /start_profile and /stop_profile, runs analyse(), verifies kernel_details.csv landed, and writes a manifest the analysis skill can consume. Use for requests like "采集 profiling", "torch profiler 跑一个 case", "采一份 profile 出来", "采 profiling 给我分析". Do not use for pure performance benchmarking, HBM/memory profiling, or for analysing already-collected profiling data (that is the analysis skill's job).
Ensure a ready remote runtime runs the exact current local workspace state before any remote smoke, service launch, or benchmark. Use automatically immediately before remote execution when direct local -> container SSH already works and local uncommitted changes must be reflected remotely. Do not use for initial machine attach, generic Git topology work, or unrelated local-only coding.
Initialize this workspace after clone. Use for requests like “初始化仓库”, “配置 gh / GitHub 登录”, “初始化子模块”, or “把 vllm / vllm-ascend remotes 改成我的 fork”. Do not use for ordinary coding, serving, benchmarking, or unrelated Git tasks.
Start, check, or stop a single-node vLLM Ascend online service on a workspace-managed ready remote container. Use for requests like "拉服务", "在远端起个服务", "重启服务", "看服务状态", "停掉服务". Do not use for machine attach, environment bootstrap, code sync, benchmark orchestration, or offline inference.
Download, resume, status-check, and SHA256-verify ModelScope model weights. Use for $modelscope download/status/verify/check, Chinese requests to 下载/续传/补全/查看进度/校验 ModelScope 权重, and tasks that need durable background ModelScope downloads under explicit local directories.