Skip to main content

Skills

turing-cli skills installs the Agent Skills that Turing publishes into the agents you use on this machine, and handles later updates and removal. It requires turing-cli 0.4.0 or later.

Available Skills​

SkillPurpose
turing-memoryRecall and Remember the same personal long-term memory across agents. See Memory
turing-operationsQuery usage, cost, and tokens on the Turing platform. See Operations
turing-artifactsCreate, edit, render, and share Artifacts on the Turing platform
everything-2-markdownConvert PDF / Word / Excel / PowerPoint into Markdown. See Documents

Run turing-cli skills list to see every Skill in the current release channel.

Install​

Run in a terminal:

turing-cli skills install

The CLI asks you to choose Skills, a scope, and agents in that order, lists the directories it will write, and then asks for confirmation:

  • Scope defaults to the current user, which applies to every project. You can choose the current project instead; the working directory where you run the command is the project directory.
  • Agent list preselects the agents it detects, and you can pick any supported agent by hand. Only agents with a directory in the chosen scope are listed.

You can also pass everything as flags:

turing-cli skills install --skill turing-operations --agent claude-code --agent opencode --global --yes
turing-cli skills install --skill '*' --agent codex --project --yes
FlagDescription
--skillSkill name; repeatable. '*' means all and must be used on its own
--agentAgent name; repeatable. See Supported agents for values
-g, --globalInstall for the current user
-p, --projectInstall into the current project
-y, --yesAsk nothing; see the note below
--catalog-urlSkill catalog URL. Defaults to the CLI's own release channel
Running from an agent or a script

--yes is the same as pressing Enter at every prompt: it uses the current-user scope and the detected agents, and skips confirmation. It fails if no agent is detected. A command with --yes never stops to wait for input, in any environment, so an agent running it for you should always pass --yes. Skills have no default, so --skill is still required.

A non-interactive run without --yes, such as in CI, must pass --skill, --agent, and either --global or --project.

Supported agents​

Agent--agentCurrent-user directoryProject directory
Claude Codeclaude-code~/.claude/skills, follows CLAUDE_CONFIG_DIR.claude/skills
OpenCodeopencode~/.config/opencode/skills, follows XDG_CONFIG_HOME.agents/skills
Codexcodex~/.agents/skills.agents/skills
CodeBuddycodebuddy~/.codebuddy/skills.codebuddy/skills
OpenClawopenclaw~/.openclaw/skillsNot supported
  • When several agents share a directory, such as .agents/skills in a project, it is written only once.
  • An agent counts as detected only when its configuration directory holds more than a skills/ directory.
  • --project is rejected in your home directory, where it would be the same set of directories as the current-user scope.
  • Codex reads user Skills from ~/.agents/skills. $CODEX_HOME/skills is a deprecated location Codex keeps only for compatibility, so CODEX_HOME is used only to detect whether Codex is installed.
  • OpenClaw reads project Skills from the workspace set in its own configuration, not the current directory, so it does not support --project. It loads ~/.agents/skills ahead of its own directory, and the copy there wins on a name clash. Installing for OpenClaw therefore also updates the copy there when it is an old installer's copy or an unmodified managed one; nothing is created if there is none, and an edited copy is left alone. status and update for OpenClaw include that directory.

Status and updates​

turing-cli skills status
turing-cli skills update
  • status lists the Skills that turing-cli manages, for the current user and the current project. Add --global or --project to show one scope, or --agent to show specific agents.
  • update updates the current-user scope by default; update --project updates the current project. It downloads only when a SHA-256 differs from the catalog. One failing Skill does not block the others, and all failures are reported at the end.
  • Each Skill keeps the release channel it was installed from. To switch channels, pass --catalog-url explicitly.

Remove​

turing-cli skills remove --skill turing-operations --agent opencode --global --yes

remove deletes only copies that turing-cli installed and that have no local changes.

  • remove --yes without --agent covers every agent, as status and update do. A same-name directory left by another tool, with no receipt, such as a copy npx skills keeps in ~/.agents/skills, is skipped and reported.
  • If such a directory is under an agent you named with --agent or picked in the picker, the whole request fails and nothing is removed. A copy with a broken receipt, or a symlink to a managed copy, also fails the request.
  • ~/.agents/skills is Codex's directory, so remove --agent openclaw does not delete the copy there. It tells you to add --agent codex to remove it as well.

Receipts and local changes​

Every installed directory contains a receipt, .turing-cli-skill.json, that records the source catalog, the version, and the SHA-256 of each file. turing-cli updates or removes only directories it installed and recorded:

  • It refuses to overwrite an existing Skill of the same name that turing-cli did not install.
  • A copy with local changes is never overwritten by update or remove. Files such as .DS_Store and __pycache__ do not count as local changes.

Release channels​

Skills are published with the CLI binaries. Stable releases and development snapshots each have their own versions and latest pointer:

  • A stable CLI reads https://live-public-aigc-resources.s3.cn-north-1.amazonaws.com.cn/turing/turing-cli/latest/skills/manifest.json by default.
  • A development snapshot reads the same path under turing-cli-dev by default.

Before installing, the CLI verifies each complete Skill package against the SHA-256 in the catalog, and it refuses to follow a redirect to a non-HTTPS address. Upgrading the CLI itself is described in Upgrade & Version; it is separate from updating Skills.

Memory​

turing-memory also needs a login and a Space configuration. After installing it with skills install, run:

turing-cli memory login
turing-cli memory install --config ~/turing-agent-memory.json --config-only
turing-cli memory doctor

--config-only imports the configuration without touching agent directories. You can also skip skills install and run memory install --config directly; it uses the same agent picker. See Memory.

Copies left by an older memory install​

Before 0.4.0, memory install always wrote turing-memory into ~/.claude/skills, ~/.codex/skills, ~/.codebuddy/skills, ~/.openclaw/skills, and ~/.agents/skills, without a receipt.

  • Only a copy whose content matches, byte for byte, a body that older release shipped is recognized as such a copy. A same-name directory with any other content is treated as not installed by turing-cli and is never overwritten or deleted.
  • Installing turing-memory again for that agent adopts the old copy. status shows it as legacy with a command you can run as-is to adopt it, and update and remove skip it.
  • Installing for Codex adopts the old copy in ~/.agents/skills. Whenever an install writes the new copy into ~/.agents/skills, for Codex or when updating it for OpenClaw, the old copy in the deprecated ~/.codex/skills is deleted, so Codex does not load two copies. The interactive confirmation lists that directory before it is deleted.
Self-check