Lekha Poth CLI (lp) v2

Production Typer + Pydantic client for every Lekha Poth HTTP endpoint, generated from backend/src/app/api on main (kept in sync — currently 140 routes as of product v0.16.2 / 6d2c833; regenerate with generate_spec.py after any route change, guarded in CI by backend/tests/unit/test_agent_spec_current.py).

Replaces the old stdlib scripts/agent/lp.py reader CLI.

Install

Published on PyPI as lekha-poth-cli. No checkout needed:

uv tool install lekha-poth-cli      # → $HOME/.local/bin/lp
uv tool upgrade lekha-poth-cli      # later, to pick up new routes and flags
$HOME/.local/bin/lp --version

On macOS /usr/bin/lp is the print spooler and is on PATH, so call the Lekha Poth CLI by its absolute path ($HOME/.local/bin/lp), or use the alias entry point lekha-poth, which nothing else claims. To try it without installing: uvx --from lekha-poth-cli lp --help.

An agent skill that teaches an LLM agent to drive this CLI ships in the sdist under skills/lp-cli/ (and in the repository at scripts/agent/lpcli/skills/lp-cli/); its README.md says where to put it.

Developing against a checkout

cd scripts/agent/lpcli
uv venv .venv && uv pip install --python .venv/bin/python -e ".[dev]"
.venv/bin/python -m lekha_poth_cli --version
# or, as a tool that tracks the working tree:  uv tool install -e .

Releasing

Actions → Release (lekha-poth-cli) → Run workflow (bump minor by default, or an exact version). It stamps every copy of the contract version (scripts/ci/stamp-lpcli-version.py), runs the tests, uv builds, commits and tags lpcli-vX.Y.Z on main, then uv publishes to PyPI. See docs/CICD.md → "Releasing the agent CLI".

Configure (once)

lp config set --api-base "$LEKHA_POTH_API_BASE" \
              --site-url "$LEKHA_POTH_SITE_URL" \
              --api-key lpak_…
# hosted production example:
#   --api-base https://api.lekhapoth.com/api/v1 --site-url https://lekhapoth.com
lp config set-defaults --output pretty --default-size 20
lp status
lp whoami

Config file: ~/.config/lekha-poth/config.json (mode 0600). Existing keys (api_base, api_key, site_url, key_name, created_at) are kept.

Env overrides: LEKHA_POTH_API_BASE, LEKHA_POTH_API_KEY, LEKHA_POTH_SITE_URL, LEKHA_POTH_CONFIG.

Coverage

lp endpoints              # every route, from endpoints.json
lp endpoints --admin
lp --help
lp admin items --help
lp items get --help       # every query/body field is a flag

Raw escape hatch:

lp api request GET /items --param size=5 --param kind=text
lp api request GET /health --root

Staff media (initiate → direct PUT → complete → poll):

lp upload ./ep.mp4 --visibility public --role original --item-id UUID --wait

Iron rules (from the live API)

• Wire JSON is camelCase except pagination: {size, next_cursor, has_next} (never hasNext, no total on public lists).
• Errors: {"error": {code, message, details, request_id}} — branch on error.code.
• coverUrl is stable or null. assets[].url may be presigned (~15 min); match assets by role.
• Legacy /chapters and /series exist only when APP_LEGACY_API_ENABLED.
• Never store presigned URLs. Never put lpak_ in git, skills, or Hermes memory.

Regenerating the route catalogue

src/lekha_poth_cli/endpoints.json drives every command, and it is derived from the live FastAPI app — there is no intermediate artefact to keep in sync:

cd backend
uv run python ../scripts/agent/lpcli/scripts/generate_spec.py          # rewrite it
uv run python ../scripts/agent/lpcli/scripts/generate_spec.py --check  # verify only

--check writes nothing and exits non-zero when the tracked file is stale. backend/tests/unit/test_agent_spec_current.py runs the same comparison in CI, so adding a route to the API without regenerating fails the build.

A new route stops the generator with an error rather than inventing a command name for it: the id is what a person types, so it is a decision. Add an entry to NEW_IDENTITY in the generator and re-run. Existing ids, auth values and hand-shortened flag names (--status for the parameter status_filter) are preserved on every run — they are the public CLI surface.

Regenerating covers the routes. docs/agent/*, agent/SKILL.md, public/llms.txt and public/.well-known/agent.json are still updated by hand.