| name | benchos-cli |
| description | BenchOS CLI command reference for workspace, memory, API, and skill-management operations. Use when generating, reviewing, or troubleshooting benchos-cli command usage. |
| license | Proprietary |
| compatibility | Requires BenchOS agent runtime with `labbook_cli` tool and `benchos-cli` installed. |
| metadata | {"author":"benchos","version":"2"} |
| allowed-tools | labbook_cli |
BenchOS CLI
Use this skill when you need exact benchos-cli command patterns.
Quick Rules
- Run commands through
labbook_cli with JSON args, e.g. {"args":["workspace","ls","."]}.
- Treat workspace operations as sandboxed to
BENCHOS_WORKSPACE_DIR (default /workspace).
- API calls automatically use
LABBOOK_AGENT_TOKEN or BENCHOS_AGENT_TOKEN.
- Prefer
search for discovery and list-style lookups; search entries returns titles only by default.
- Use
api --raw when you need the full response payload instead of the compact default.
- User skills created by CLI are user-local by default and require explicit share request + admin approval.
Command Catalog
For full command patterns, read references/COMMANDS.md.
High-Value Workflows
- Verify runtime identity and connectivity before multi-step tasks.
- Use
search entries first to shortlist by title, then fetch the full object only after you know the relevant entry.
- Prefer
workspace commands for file I/O over ad-hoc shell behavior.
- For binary uploads, use
api --file and include --form fields when endpoint requires multipart metadata.
- For script execution, install a curated capability pack first and then use
exec.
- Do not install new packages inside the persistent sandbox; request a new curated capability pack when a skill needs additional Python, Node, or OS dependencies.
- For external websites, use
web fetch; do not assume arbitrary outbound network access.
- For skill publishing workflow:
skills create for user-local draft- validate behavior on real tasks
skills request-share for admin review
