Agent-Skills.md

Agent Skills: ZeP CLI

Use when operating the ZeP (Ze Experiments Platform) through its first-party `zep` CLI — authenticating (local or "Sign in with Zerg" SSO), minting/listing/revoking scoped API tokens, inspecting experiments/runs/datasets/backends, submitting runs, and reading the integration API discovery doc.

UncategorizedID: idanbeck/claude-skills/zep-cli

Repository

idanbeck/claude-skills
idanbeck
123

Install this agent skill to your local

pnpm dlx add-skill https://github.com/idanbeck/claude-skills/tree/HEAD/zep-cli

Skill Files

Browse the full folder contents for zep-cli.

Download Skill

Loading file tree…

zep-cli/SKILL.md

Skill Metadata

Name
zep-cli
Description
Use when operating the ZeP (Ze Experiments Platform) through its first-party `zep` CLI — authenticating (local or "Sign in with Zerg" SSO), minting/listing/revoking scoped API tokens, inspecting experiments/runs/datasets/backends, submitting runs, and reading the integration API discovery doc.

ZeP CLI

Use the zep CLI instead of scraping the ZeP web UI when you need to inspect or operate experiments, runs, datasets, scrapers, backends, storage, or API tokens.

Defaults

  • Run from /Users/idanbeck/zerg-ze/ze (invoke with poetry run zep ..., or zep ... if installed on PATH).
  • Config is ~/.zep/config.json; environment overrides are ZEP_API_URL, ZEP_TOKEN, ZEP_API_KEY, ZEP_WORKSPACE_ID, ZEP_CONFIG_PATH (env wins over the saved file).
  • Default API base is http://localhost:8080. Set once with zep config set-api-url <url>.
  • Prefer a scoped zep_ API token for automation; use zep auth login for interactive/operator tasks.
  • Many commands accept --json or already print JSON; prefer JSON for scripting.

Authenticate

Local credentials:

zep auth login you@example.com           # prompts for password
zep auth whoami

Sign in with Zerg (SSO against zergai.com, if enabled on the server):

zep auth login you@zergai.com --sso       # prompts for your zergai.com password

Mint a scoped automation token

zep auth token create \
  --name agent \
  --workspace 1 \
  --scopes experiment:read,run:read,dataset:read,run:write \
  --expires-in-days 30
# The raw zep_... token is printed ONCE. Store it, e.g.:
export ZEP_TOKEN=zep_xxxxxxxx

List / revoke tokens:

zep auth token list
zep auth token revoke <id>

Inspect & operate

zep discovery                 # integration API capabilities (version, scopes, resources)
zep experiments list
zep runs list
zep runs submit <experiment> --backend modal
zep datasets list
zep backends list

Target a specific workspace with --workspace <id> or ZEP_WORKSPACE_ID. A workspace-pinned token ignores other workspace selections.

Scopes

Scopes are <resource>:<action> (e.g. run:read, experiment:write). A :write scope implies :read. Run zep discovery to see the authoritative list and each resource's required scopes. A token can never be minted with broader scopes than its creator holds.

Safety

  • Do NOT submit runs, cancel, or delete unless the user asks.
  • Show the target workspace before any mutating command.
  • Never echo a raw zep_... token into logs or chat; prefer export ZEP_TOKEN=... set out-of-band.
  • For agent automation, always use a narrowly-scoped zep_ token, never an operator session token.