界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
效率与协作 · SKILL.md

cli-and-automation

cli-and-automation

【效率与协作】Day-to-day OpenSquilla CLI automation for chat, agent runs, sessions, memory, cron, cost, diagnostics, replay, migration, recovery, sandbox posture, bundle, dist, reset, init, and uninstall.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: cli-and-automation description: "Day-to-day OpenSquilla CLI automation for chat, agent runs, sessions, memory, cron, cost, diagnostics, replay, migration, recovery, sandbox posture, bundle, dist, reset, init, and uninstall." metadata: disco-role: operating disable-model-invocation: true license: Apache 2.0

CLI and Automation

Use this sub-skill as the command hub for an existing OpenSquilla user who has already installed the package and is asking how to run, automate, inspect, or clean up ordinary CLI workflows. The evidence for this skill was checked against OpenSquilla 0.5.3 in a CPU-only inspection build where the opensquilla entry point and assigned command help surfaces loaded successfully.

Load This When

  • The user asks for opensquilla chat, agent, code-task, sessions, memory, agents, cron, cost, diagnostics, replay, bundle, dist, migrate, recovery, sandbox, reset, init, or uninstall usage.
  • The user wants non-interactive CLI patterns with --json, --timeout, workspace controls, transcripts, usage files, progress events, or bounded retries.
  • The user needs a safe sequence for session export/resume, memory indexing/repair/flush, scheduled jobs, cost inspection, diagnostics/replay, migration preview/apply, recovery, or data removal.
  • The user is confused about gateway-backed commands versus local/offline commands after first setup.

Route Elsewhere

  • First-run install, onboarding, gateway lifecycle, Web UI launch, or doctor readiness basics: setup-and-gateway.
  • Provider credentials, model selection, router modes, model catalog, search providers, or raw config precedence: configuration-and-routing.
  • Messaging channels, pairings, channel delivery, and MCP bridge setup: channels-and-integrations.
  • Skill catalog, skill install/update/publish, or meta-skill runs: skills-and-meta.
  • Terminal UI renderer selection, OpenTUI behavior, desktop shell startup, or Web UI presentation details: tui-and-desktop.

Operating Rules

  1. Identify whether the command is gateway-backed before prescribing it. Sessions, most memory inspection commands, cron, cost, diagnostics, and reset need a reachable gateway; agent, chat --standalone, code-task, replay, dist, bundle, migrate, recovery, sandbox, init, and uninstall are local/offline or manage their own process as documented in the bundled references.
  2. Prefer --json and explicit confirmation flags for scripts, but never hide destructive consequences. uninstall and cron remove/run have confirmation behavior; code-task solve requires --yes for non-interactive trusted-host use; migrate is dry-run until --apply.
  3. For automation that can touch files, set an explicit --workspace; add --workspace-strict, --workspace-lockdown, and --scratch-dir when containment matters. Use per-run transcript, usage, event-stream, and session DB paths when jobs must be reproducible.
  4. Keep sharing surfaces redacted. Session exports, diagnostics bundles, replay output, raw diagnostics, and bundle --include-content can contain prompts, tool results, private paths, secrets, or channel identifiers.
  5. Do not assume the original repository checkout exists at runtime. Use only the command patterns and local references bundled in this generated skill.

Reference Map

  • Command map: command groups, gateway requirements, JSON support, and common examples.
  • Automation patterns: non-interactive agent, code-task, cron, event-stream, workspace, and parallel-run patterns.
  • State and data workflows: sessions, memory, diagnostics, cost, migration, recovery, bundle/dist, reset, init, sandbox, and uninstall workflows.
  • Troubleshooting: gateway dependency failures, session export risk, stale memory index/repair, migration conflicts, uninstall purge safety, sandbox posture confusion, optional dependency gaps, and live-script boundaries.

Quick Triage

  • "Run one prompt and return machine-readable output" → opensquilla agent --json -m "..."; see automation patterns.
  • "Continue or export a prior conversation" → opensquilla sessions ...; see state workflows.
  • "Find or repair durable memory" → opensquilla memory ...; see state workflows.
  • "Schedule repeated work" → opensquilla cron ...; see automation patterns.
  • "Debug a surprising or costly turn" → diagnostics, cost, replay, and bundle; see state workflows.
  • "Move state from another runtime or recover profile data" → migrate or recovery; see state workflows.
  • "Remove OpenSquilla or reset data" → start with uninstall --dry-run; see state workflows.
我的收藏