| name | odk-run |
| description | Run custom Make targets or ODK tools that have no dedicated skill. Use when the user wants a one-off make target (other than test, prepare_release, refresh-imports, update_repo, odkversion) or to run Konclude, jena, relation-graph, jinjanate, OORT, souffle, dicer-cli, yq-mf, or sssom-cli. Do not use for robot, owltools, dosdp-tools, fastobo-validator, riot, sparql, runoak, or sssom—use those skills instead. |
ODK Run Skill
Use this skill when the user needs to run a command that is not covered by a dedicated ODK skill: either a custom Make target in the ontology dir, or one of the other ODK tools (Konclude, Apache Jena, relation-graph, jinjanate, ontology-release-runner, souffle, dicer-cli, yq-mf, sssom-cli). In this Ontology Builder repo the container mounts the project directory (workspace root or projects/<slug>) as /work; use paths relative to that root and make -C ontology <target> (or the Makefile path) as appropriate. For ROBOT, owltools, dosdp-tools, fastobo-validator, riot, sparql, runoak, or sssom, use the corresponding odk-* skill.
When to Use This Skill (by outcome)
- User wants a one-off or custom Make target that is not
test, prepare_release, refresh-imports, etc. → use odk_run with command: make -C ontology <target> (or make -C <make_path> <target>) (paths relative to the mounted project root, e.g. projects/<project_dir>/ or workspace root).
- User wants to run Konclude, jena, relation-graph, jinjanate, OORT, souffle, dicer-cli, yq-mf, or sssom-cli → use odk_run with
command set to the full invocation (paths relative to the mounted project root).
What the Tool Does and How It Works
What it does: Runs an arbitrary single command in the ODK environment. Use only for custom make or the tools listed above; for robot, owltools, dosdp-tools, riot, sparql, runoak, sssom, and fastobo-validator use the dedicated skills so the agent loads only what it needs.
How it works: Pass the full command string to odk_run via the command parameter. Optionally pass project_dir (e.g. projects/pizza or projects/owner-repo) to mount that directory at /work; relative paths are resolved from the repo root. If project_dir is omitted or empty, the current working directory is mounted. Paths in command are relative to the mounted root. Example: make -C ontology custom-target or Konclude -i ontology/edit.owl -o ontology/out.owl.
When Not to Use This Skill
- ROBOT → use odk-robot and odk_robot.
- owltools → use odk-owltools and odk_owltools.
- dosdp-tools → use odk-dosdp-tools and odk_dosdp_tools.
- fastobo-validator → use odk-fastobo-validator and odk_fastobo_validator.
- riot → use odk-riot and odk_riot.
- sparql → use odk-robot and odk_robot with
query.
- runoak → use odk-runoak and odk_runoak.
- sssom → use odk-sssom and odk_sssom.
- Standard ODK Make targets (test, prepare_release, refresh-imports, update_repo, odkversion) → use odk-make and odk_make.
- Seeding a repo → use odk-seed and odk_seed.
Tool and Arguments
| Argument | Type | Required | Description |
|---|
| command | string | yes | Full command (no robot, no owltools/dosdp-tools/riot/sparql/runoak/sssom/fastobo-validator). E.g. make -C ontology custom-target, Konclude -i edit.owl -o out.owl. Paths relative to the mounted project root. |
| project_dir | string | no | Optional. Project root to mount at /work (e.g. projects/pizza, projects/owner-repo). Relative paths resolved from repo root. Omit or leave empty to use current working directory. |
Examples (calling the tool)
- Custom make (workspace): odk_run with
command: make -C ontology some-custom-target
- Custom make (clone): odk_run with
command: make -C ontology some-custom-target, project_dir: projects/owner-repo
- Konclude: odk_run with
command: Konclude -i edit.owl -o classified.owl (optionally with project_dir if running in a project)
- Other tool: odk_run with
command: <tool> --help then the full invocation the user needs
Tool Requirement
This skill requires: odk_run
After learning this skill, call setup_tools(skills: ["odk-run"]) to activate the tool, then use call_tool(name: "odk_run", data: {"command": "..."}) to invoke it.
