Agent Skills¶

langgraph-xai publishes one portable Agent Skill that teaches coding agents how to integrate, configure, test, and debug langgraph-xai in an application. The skill is guidance for agents; it is not part of the Python package and does not change how langgraph-xai is installed.

Component	Location
langgraph-xai Python package	`src/langgraph_xai/`
The Agent Skill	`langgraph-xai-skills/skills/langgraph-xai/`
Its instructions	`SKILL.md`

What the skill contains¶

The skill follows the Agent Skills specification. Agents read SKILL.md first and open the other files only when a task needs them:

SKILL.md: when to use the skill, the core workflow, and the rules an integration follows. Its frontmatter has name, description, license, compatibility, and metadata.version, the langgraph-xai release it describes.
references/API.md: public imports, runtime methods, settings, records, and provider contracts.
references/RECIPES.md: complete patterns for instrumenting graphs and agents, human-in-the-loop, disclosure policies, storage, tracing backends, failure modes, and LLM-phrased explanations.
references/TROUBLESHOOTING.md: symptoms, causes, and fixes.
scripts/verify_setup.py: an offline check that the project's environment has compatible versions, and that a small instrumented graph records, explains, and links runs correctly.
assets/test_disclosure_policy.py: a pytest template that tests what each audience may and may not see.

The repository validates the skill on every change: the reference validator, link checks, the setup script, and the template run in its test suite. There is no separate Claude, Codex, Cursor, or Copilot copy of the skill.

Install from skills.sh¶

The skills CLI installs skills from a GitHub source. Install the langgraph-xai skill directory directly:

npx skills add https://github.com/smuniharish/langgraph-xai/tree/master/langgraph-xai-skills/skills/langgraph-xai

This is the portable installation route. Follow the CLI's current target selection prompts, then verify that it placed the langgraph-xai folder in the target agent's supported skills directory. The shorthand npx skills add langgraph-xai-skills is not a valid source identifier.

Install manually¶

First obtain the canonical skill directory from the repository. Copy the complete langgraph-xai directory, including SKILL.md, into one of the host-specific locations below.

Claude Code¶

Claude Code discovers standalone skills in:

Scope	Destination
Current repository	`.claude/skills/langgraph-xai/`
All local projects	`~/.claude/skills/langgraph-xai/`

Start or restart Claude Code after copying the directory. Claude can select the skill when its description matches the task, or you can invoke it with /langgraph-xai. See Claude Code Skills.

langgraph-xai does not currently ship a Claude plugin manifest or marketplace package. The standalone skill is sufficient today.

Codex¶

Codex discovers repository skills by scanning .agents/skills from the working directory to the repository root. Copy the directory to:

Scope	Destination
Current repository	`.agents/skills/langgraph-xai/`
All local projects	`~/.agents/skills/langgraph-xai/`

Codex detects changes automatically; restart it if the skill does not appear. Invoke it explicitly with $langgraph-xai or use /skills to inspect available skills. See ChatGPT and Codex Skills.

Cursor¶

Cursor supports the standard .agents/skills locations, which makes the Codex layout above portable. It also supports Cursor-specific locations:

Scope	Destination
Current repository	`.agents/skills/langgraph-xai/` or `.cursor/skills/langgraph-xai/`
All local projects	`~/.agents/skills/langgraph-xai/` or `~/.cursor/skills/langgraph-xai/`

Restart Cursor after copying the directory. In Agent chat, type / and select langgraph-xai to attach it to a message; Cursor can also activate it from its description. See Cursor Agent Skills.

GitHub Copilot¶

GitHub Copilot supports the standard .agents/skills layout, and also the following project and personal locations:

Scope	Destination
Current repository	`.agents/skills/langgraph-xai/`, `.github/skills/langgraph-xai/`, or `.claude/skills/langgraph-xai/`
All local projects	`~/.agents/skills/langgraph-xai/` or `~/.copilot/skills/langgraph-xai/`

For Copilot CLI, start a new session or run /skills reload, then verify the skill with /skills info langgraph-xai. You can explicitly request it in a prompt using /langgraph-xai. See Adding agent skills for GitHub Copilot CLI.

Other Agent Skills-compatible hosts¶

Use the host's documented skill directory and copy the complete langgraph-xai folder there. The canonical skill relies only on the standard SKILL.md frontmatter, so it does not require a host-specific adapter.

If a host needs package metadata, a registry entry, or a plugin manifest, follow that host's current official documentation and add only a thin adapter that points to this canonical skill. Do not duplicate the skill instructions.

Update and verify¶

To update a manual installation, replace the complete installed langgraph-xai folder with the latest directory from the repository, then restart or reload the host.

After installation, confirm all of the following:

The directory name is langgraph-xai, and it contains SKILL.md, references/, scripts/, and assets/.
The host lists langgraph-xai as an available skill, if it exposes a skill listing command or UI.

From your project's Python environment, the setup check passes:

python <skills directory>/langgraph-xai/scripts/verify_setup.py

PASS  Python 3.14.7 (3.12 or newer required)
PASS  langgraph-xai 1.0.0 (1.x expected by this skill)
PASS  langgraph 1.2.12 (>=1.2.12,<2 required)
PASS  langchain-core 1.6.6 (>=1.6.6,<2 required)
INFO  optional integrations: langsmith 0.14.4, langfuse 4.16.0, opentelemetry-sdk 1.45.0, langchain-openai 1.6.7
PASS  instrumented graph recorded nodes ['score', 'route'], 2 state changes, 1 evidence, 1 decision
PASS  auditor explanation: The routing decision selected 'REVIEW'.
PASS  end-user explanation withholds the score and announces it
PASS  no instrumentation errors (0 recorded)
PASS  interrupt and resume recorded and linked through the checkpoint
All checks passed.

A task about LangGraph explainability, provenance, evidence, decisions, disclosure policy, or instrumentation activates the skill, or you can invoke it explicitly.

See the distribution's README and validation process for maintenance details.