Installing the Torc Agent Skill
Torc ships an agent skill that teaches AI coding assistants (Claude Code, Codex, Copilot, and other
tools that support the Agent Skills format) how to author, run, inspect,
and debug Torc workflows. The skill lives in the repository at
skills/torc/.
Skill vs. MCP Server
The skill and the MCP server are complementary:
- Skill: instructions and reference material. The assistant learns Torc's spec format, execution
modes, Slurm sizing, recovery steps, and which
torcCLI commands to run. No extra binary or server configuration is required. - MCP server: tools the assistant calls directly against the Torc API.
The skill works on its own through the torc CLI, and it uses the MCP tools when they are
configured.
Install with the skills CLI
The skills CLI installs skills from GitHub into any
supported agent:
# Install into the current project
npx skills add NatLabRockies/torc --skill torc
# Install for your user account (all projects)
npx skills add NatLabRockies/torc --skill torc -g
The CLI prompts for which agents to install into. Rerun the command to pick up a newer version of the skill.
Install Manually
Copy the skills/torc directory into your agent's skills directory. For Claude Code:
git clone --depth 1 https://github.com/NatLabRockies/torc.git /tmp/torc
# User-level (all projects)
mkdir -p ~/.claude/skills
cp -r /tmp/torc/skills/torc ~/.claude/skills/
# Or project-level
mkdir -p .claude/skills
cp -r /tmp/torc/skills/torc .claude/skills/
Other agents use their own directories (for example, .agents/skills/ or ~/.agents/skills/); see
your agent's documentation.
Verify
Start a new session and ask something Torc-specific, such as "Write a torc workflow spec that runs
10 parameterized jobs on Slurm." In Claude Code, the skill also appears in the list shown by
/skills and can be invoked directly with /torc.
See Also
- Configuring AI Assistants - Set up the Torc MCP server
- AI-Assisted Workflow Management - Using AI for workflow management