feat: tests command group (add/delete/push/pull + auto-discovery)

[PaulAsjes]

What

Ports v0's src/tests/ as the tests command group — the last of the three entity groups, completing v0 parity for agents/tools/tests.

• tests add <name> [--template …] — build from a template, create in ElevenLabs, write test_configs/<name>.json, register in tests.json. All four v0 templates ported: basic-llm, tool, conversation-flow, customer-service.
• tests push [--test <id>] [--config-dir …] [--dry-run] — force-override create/update.
• tests pull [--test <id>] [--output-dir] [--update] [--all] [--dry-run] — create/update/skip planning, same shape as agents/tools.
• tests delete <id> / --all
• tests templates list

The interesting bit: push-time auto-discovery

tests push scans --config-dir recursively for .json files that aren't yet in tests.json and structurally look like tests (a chat_history array or a success_condition string), registers them, and warns on non-test JSON. So you can drop a config into test_configs/ and push without hand-editing the index. Ports v0's registerDiscoveredTestConfigs/isTestConfigFile, including recursive walk and posix-normalized paths for portable index entries.

Verified end-to-end offline:

$ tests push --dry-run     # after dropping 3 files into test_configs/
Warning: Skipping non-test JSON file test_configs/not-a-test.json
Discovered 2 test config(s) in test_configs      # incl. a nested subdir
Greeting: Will push (force override)
[DRY RUN] Would create test: Greeting
Deep: Will push (force override)
[DRY RUN] Would create test: Deep

…and re-running with one already tracked discovers only the other, correctly showing update for the tracked one and create for the new one.

Also

• api.rs: tests API against v1/convai/agent-testing via the raw-JSON layer (lossless round-trip); list paginates to completion.
• project.rs: discover_json_files() helper (recursive, sorted, posix-normalized).

Verified

cargo build green; 19 workflow unit tests pass (5 new: all templates build, unknown-template error, tool template shape, multi-turn flow, and the test-config shape classifier). Offline smoke tests cover discovery, the --test filter, and template validation. (Live create/update/pull needs an API key.)

Remaining for v1

residency + components add, then release readiness (port remaining v0 unit tests, E2E smoke, docs, tag 1.0.0).

🤖 Generated with Claude Code