Skip to main content

​
Demo use cases

The second version of the demo scripts. The first, scripts/walkthrough.sh, was one checklist under set -e: the first failing line ended it, and there was no way to run step six without steps one to five. This version is twelve units under scripts/demo/, one script per use case, each runnable alone and each ending in a verdict.
The scripts sit on the CLI integration harness under scripts/test-harness and drive the real crewship binary; nothing here reads a database or calls the API by hand.

​
The use cases

Needs is what a use case cannot run without. model is an ACTIVE model-provider credential in the workspace; github is SEED_GITHUB_TOKEN at seed time, or a bound GITHUB credential. A use case whose need is missing does not pretend: it exits with a reason and the runner prints SKIP for it. A green row means the use case ran and every assertion held.

​
Reading a run

Each use case prints numbered steps, the commands it runs with their output, a ✓ or ✗ per assertion, and a 👀 line naming where to look in the web UI right now — the inbox, the agent, the Page. It ends with its own summary. The runner then prints one table:
Exit status is 1 when any use case failed. A SKIP never fails the run; the reason is in the NOTE column.

​
Before you start

A seeded install and a login: crewship seed prints the account. The use cases with a model need the seed’s model credential to be ACTIVE; eval-tiers needs crewship seed --with-evals. Point the scripts at the same server the browser shows (CREWSHIP_PROFILE or CREWSHIP_SERVER) and, when the UI lives on another host than the API, set DEMO_WEB_URL so the 👀 lines are right. The packs with a GitHub token (10, 11) need the token at seed time — see Demo packs → Credentials. Without it they SKIP, and so does seed verify.

​
Writing one

A use case is a bash script scripts/demo/uc-NN-<slug>.sh with a four-line header the runner lists from, source "$HERE/lib.sh", preflight, its steps, and finish. lib.sh gives it the harness (cs, ask_agent, assert_*, poll_until, nonce) plus the demo layer: demo_step, demo_say, demo_show, demo_ui, demo_need and demo_skip_all, run_routine, demo_cleanup. There is no set -e on purpose: a failed assertion is a red line, and the audience still sees the steps after it. scripts/demo/README.md has the contract; the existing twelve are the examples.

​
Known limits

  • crewship routine schedules now runs the target routine directly and bypasses the wake gate, so use case 7 waits for a real tick instead (under 90 seconds).
  • Use cases 2 and 4 depend on an agent doing what it was asked. Each retries once, then SKIPs with the reason rather than reporting the model’s phrasing as a product failure.
  • What walkthrough.sh still shows and this does not: nothing. Its steps are use cases 1 to 9 here; step 8 (seeded cron schedules) no longer matches what the seed creates and became use case 7.