Discord E2E

Choose the smallest boundary that proves the requested behavior:

The API probe starts no Gateway, event recorder, or model. QA Lab remains the owner for product E2E: it controls its lease, temporary Gateway, provider, cancellation, and cleanup. Neither path changes app permissions or provisions credentials.

1. Choose the evidence

Read feature recipes for the changed behavior. Native fixture calls prove Discord API operations, not that a model invoked a tool. A SUT round trip additionally proves Gateway ingress and a stored public reply. Model-tool claims also require the tool trace and provider request evidence.

Human slash commands, component clicks, modals, ephemeral interactions, and human-to-bot DMs need an authorized manual client. A bot posting /status sends text, not a slash interaction. Use official bot tokens from the lease only. Discord prohibits user-token/self-bot automation; Slack's user OAuth route has no equivalent here.

2. Prepare and check the current lease

Use the dependency-ready OpenClaw checkout under test. The setup prerequisite is an existing authenticated Convex CLI with access to the published QA broker. The existing shared lease helper discovers the repository's broker binding and CI credential in memory; you do not need to copy bot tokens, guild IDs, or broker secrets. If authentication is absent, have the operator run convex login (or the already cached bunx --no-install convex login) once. Do not install a CLI or log in on the operator's behalf. Network/permission errors are not evidence that another login is needed.

For a fast check without QA Lab:

node .agents/skills/discord-e2e/scripts/bot-readiness.mjs

The check is read-only in Discord, but it acquires, renews, and releases a Convex lease. It creates no Discord objects and has no mutation mode. Read bot API readiness for the payload, permission limits, private result, and failure handling.

This probe does not verify Gateway intents, event delivery, model behavior, or all native fixture permissions. For an OpenClaw claim, continue with QA Lab:

pnpm openclaw qa discord --list-scenarios
pnpm openclaw qa discord --doctor \
  --output-dir .artifacts/qa-e2e/discord-doctor

Require a passing QA Lab doctor for this lane. It sends no fixture messages; it checks the leased identities, guild text channel, effective permissions, driver Gateway intents, and connected SUT Gateway. Every scenario repeats readiness on its own lease; a released doctor's result never qualifies a later run.

--doctor and --scenario-file default to Convex, CI role, and mock-openai. Existing explicit credential/provider flags still work; ordinary curated qa discord defaults remain unchanged. Preprovisioned private broker environment variables are an alternative to discovery, not a prerequisite.

3. Run one scoped proof

For a native lifecycle plus real Gateway reply:

pnpm openclaw qa discord \
  --scenario-file qa/scenarios/channels/discord-e2e-lifecycle.yaml \
  --output-dir .artifacts/qa-e2e/discord-lifecycle

For a changed feature, adapt that YAML using feature recipes. --scenario-file is repeatable and uses the existing QA flow schema. Set retryCount: 0 for native write scenarios. Use a new output directory for each proof. These examples are opt-in, not additions to the curated default suite.

Read runtime and recovery when changing config, restarting the SUT, choosing a real provider, diagnosing cancellation, or reconciling failed cleanup.

4. Judge and report

For the API probe, require status: ready and leaseReleased: true. An empty history list is inconclusive (exit 2): it can mean an empty channel or missing Read Message History permission. Use QA Lab doctor to distinguish those cases; do not call either one a proven permission failure. Report a ready result as bot identity and observed read-access proof, not mutation or Gateway E2E. Private result.json contains leased identities and check statuses, not tokens or history.

For QA Lab, inspect qa-suite-summary.json, qa-suite-report.md, the Gateway/provider artifacts, and each private discord-e2e-*/events.ndjson under the output directory. Join native receipts to recorder rows by message ID; use actor: sut, channel, sequence, and trigger correlation for SUT claims. A marker alone does not prove edits, reactions, typing, deletion, formatting, or a model tool call. Event logs are not visual proof; visual claims need an actual Discord client screenshot.

Completion requires the requested evidence and successful owned cleanup after Gateway stop, before lease release. Existing permissions decide whether owned threads are deleted or archived; report the recorded disposition, never call archival deletion. Preserve private evidence on failure. Report the sanitized command, exact revision, claim, relevant evidence, manual-only gaps, and cleanup outcome. Redact identities, unrelated content, private paths, and credentials before sharing artifacts.