@try-works/dsh-restart
作为 DeepSeek Harness bundle 的自唤醒主机重启：dsh_restart 工具会重启 DSH 主机、记录决策，并在主机重新监听后向每个活跃会话排队发送继续消息。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add @try-works/dsh-restart@0.1.2Restart 以 @try-works/dsh-restart 发布，当前版本为 0.1.2。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Self-waking host restart as a DeepSeek Harness bundle: a dsh_restart tool that restarts the DSH host, records the decision, and sends a continue message to every active session when the host is listening again.
dsh_restart tool — restarts the running DSH host. It derives every path and URL at runtime (nothing device-hardcoded):
fs.resolve('.'))DSH_WEB_URL, falling back to http://127.0.0.1:<webServer.port>$DSH_HOME/tmpctx.sessionQuery.listSessions()), plus optional explicit session_id + the calling agent's own session, deduplicatedreason (required) is recorded to the decision log and relayed to every woken session as a decision recorddry_run: true renders the wake script and returns the plan without touching the host.dsh-restart-wake.generated.ps1 in the logs dir) that:
pnpm dsh web,/api/session.prompt in queue mode (which maps to agent.followup — resuming each session's next turn), carrying the decision record.tools, shell, shellEnv, fs (all present in @deepseek-ai/dsh-base). sessionQuery and sessions are optional (read via ctx.get): sessionQuery is the authoritative session lookup, and sessions is its fallback. webServer is optional.dsh must be on PATH for pnpm dsh web).dsh_restart tool launches the wake script detached (cmd.exe /c powershell -File ...), so it survives the host process being killed.Add this package to the profile and its bundles list, e.g. in ~/.dsh/profiles/<name>/package.json:
{
"dependencies": {
"@try-works/dsh-restart": "^0.1.0"
},
"dsh": {
"profile": {
"bundles": [
"@deepseek-ai/dsh-base",
"@try-works/dsh-restart"
]
}
}
}
Then pnpm install in the profile and restart the running DSH process. The bundle ships its src/*.ts sources directly (DSH loads them through tsx), mirroring dsh-paper-design and dsh-recursive-mode.
Restart is required for host/bundle changes: the web client-plugin HMR only rebuilds the web shell, not host plugins.
Ask the agent to restart the host, e.g.:
Restart the DSH host because we applied plugin changes.
The agent calls dsh_restart with a reason; the tool writes the decision record, launches the wake script, and returns the plan. When the host is back, every session active at trigger time receives a continue message (queued as its next turn):
RESTART-CONTINUE [boot=abc12345] host is UP. pid=12804 | count=3 hasRecursive=true | Please continue your work. DECISION RECORD: applying plugin changes
pnpm install
pnpm run build # type-check (tsc -p tsconfig.build.json)
pnpm run test:unit # vitest unit tests (template substitution)
pnpm run verify # in-process boot + dry-run (needs a live shell/fs environment)
pnpm test # build + verify + unit
src/index.ts — bundle entry: registers the dsh_restart tool on ctx.tools, resolves runtime defaults, writes the script + decision log, launches the wake script detached.src/wake-script.ts — template loader + placeholder substitution (pure, unit-tested).assets/dsh-restart-wake.ps1.template — the wake-script template (placeholder contract in src/wake-script.ts).tests/ — vitest unit tests.scripts/verify-restart.mjs — in-process boot + dry-run verification.cordis.patch.yml — loader patch inserting the plugin into a cordis:group realm.$pid is a read-only automatic variable — the template's kill loop uses $procId and must never assign $pid.dsh-restart-host.out.log, dsh-restart-err.log) so the wake log stays clean.recursive/* event types are unknown to a resumed harness; keep them registered in packages/core/session/src/known-event-types.ts (see the original README.md in the workspace) — don't run gen-persistence-catalog without re-adding them.MIT