@evg-olympus/cli

Operator CLI for Olympus — maintain content (skills, tools, agents, approvals) and read fine-grained spend analytics — over the Admin Console API. Anyone on the operator allowlist can use it; the same Google SSO + allowlist that gates the admin console gates the CLI.

npm install -g @evg-olympus/cli
olympus login                 # opens the browser, approve on the /cli-auth page
olympus whoami

Credentials live in ~/.config/olympus/credentials.json (mode 0600), one entry per --profile so prod and staging never mix.

By default the CLI targets the production backend, https://forgeolympus.dipbit.xyz (Kubernetes; the API is served at the host root, with no /olympus path prefix). Point it at a local or staging backend with olympus login --api https://… or by exporting OLYMPUS_API. Profiles still pinned to the retired ECS gateway are rewritten to the current production base on first use, so an upgrade needs no manual edit.

Commands

| Area | Command | Notes | |---|---|---| | Auth | login / whoami / logout | device-authorization grant; auto-refresh | | Skills | skills list | catalog overview | | | skills lint <dir> | validate against the exact backend rules, locally | | | skills pack <dir> | build the upload zip | | | skills push <dir> | preview a diff, then publish (--dry-run, --yes) | | Approvals | approve <skill> [--standing] | approve a runnable bundle's capabilities | | | revoke <skill> / capabilities <skill> | | | | approvals list / approvals decide <id> | the runtime tool-approval queue | | Tools/MCP | tools list / mcp list / mcp test <name> | | | Agents | agent list / show <id> | show reads the summary view — the soul is listed, not inlined | | | agent set-skills <id> <skill...> / add-skill | server validates every name | | Souls | agent rules <id> | every rule the agent references, with its size | | | agent rule <id> <shared\|private> <name> | one rule's markdown — a window by default, --offset/--limit, --all, --out FILE | | | agent prompt <id> | the assembled system prompt, same windowing | | | agent rule-put <id> <scope> <name> | write one rule from --content/--file | | Conversations | conversation model <id> / set-model <id> <ref\|reset> | switch which model a thread's turns run on (per-(user, agent) preference, same row as /model) | | Spend | spend kpis / breakdown / series / purpose / model / calls | read the LLM-call ledger; window by --days or --start/--end; --json, --by/--split-by (breakdown), --interval (series), filters + --limit/--offset (calls) |

Every write is two-step (preview → confirm) and honours --dry-run. Auth, permissions, and audit are all enforced by the backend — the CLI never decides access itself.

Strictness

skills lint is a faithful port of the backend's skill_from_zip validation (file-count/size limits, path safety, frontmatter, parameter and runnable-manifest rules), verified to produce the same parsed tree as the server. A skill that lints clean here will not be rejected on push.

Bundles

bundles/agent-assembler and bundles/admin-ops are first-party skills that let a capable agent assemble a working agent for a non-technical person — strictly, from existing and approved capabilities only, stating plainly when a need must go to an administrator. Install them with olympus skills push bundles/<name> and assign with olympus agent set-skills.

Development

npm install
npm run build      # tsc → dist/
npm test           # vitest
npm run typecheck