rollcall

Rollcall

End-to-end driver parity fixture for Clawdapus.

This fixture reuses one Discord bot identity across seven runtime families, so the spike now materializes one runtime at a time rather than pretending to be a concurrent social-topology test. Each subtest wires one runtime through cllama passthrough, exposes clawdash, posts a Discord mention, and verifies that the runtime replies identifying itself.

What It Covers

• openclaw
• nullclaw
• microclaw
• nanoclaw
• nanobot
• picoclaw
• hermes

Each service shares the same Discord bot token. The distinction between agents comes from their AGENTS.md contracts and per-service runtime configuration. Because the identity is shared, this fixture is for sequential runtime conformance only. It is not a valid concurrent topology example.

Files

• claw-pod.yml: spike template containing the seven runtime service definitions
• agents/*/Clawfile: one Clawfile per runtime
• agents/*/AGENTS.md: minimal runtime-specific self-identification contract
• Dockerfile.*-base: local base images used by the spike test
• discord-responder.sh: helper script used by the stub runtimes

Setup

cp .env.example .env
# Edit .env with real values

Required variables:

• DISCORD_BOT_TOKEN
• DISCORD_BOT_ID
• DISCORD_GUILD_ID
• ROLLCALL_CHANNEL_ID
• DISCORD_WEBHOOK_URL
• XAI_API_KEY or OPENROUTER_API_KEY or ANTHROPIC_API_KEY

DISCORD_WEBHOOK_URL is required because the spike posts the trigger message via webhook rather than as a bot user. That avoids agents ignoring the message as self-authored bot traffic.

Run

From the repo root:

go test -tags spike -v -run TestSpikeRollCall ./cmd/claw/...

Or from this directory:

go test -tags spike -v -run TestSpikeRollCall ../../cmd/claw/...

For the integrated capability-wave path on oc-roll only:

go test -tags spike -v -run TestSpikeCapabilityWaveLive ./cmd/claw/...

Expected Result

The test should:

1. Build the base images for each runtime family if needed.
2. Build the seven rollcall agent images.
3. Materialize and run one single-service pod per runtime.
4. Wait for each runtime container to become healthy or running.
5. Post a Discord mention through the webhook for that runtime.
6. Observe seven AI-generated replies across the full test run mentioning:
   • openclaw
   • nullclaw
   • microclaw
   • nanoclaw (or Claude Agent SDK)
   • nanobot
   • picoclaw
   • hermes
7. Confirm claw audit telemetry appears for each runtime's cllama-backed turn.

The dedicated capability-wave spike additionally confirms that oc-roll can:

1. receive a real Discord trigger
2. execute a managed service tool through cllama mediation
3. record tool_trace in session history
4. emit memory_op recall and retain telemetry in the same turn

Notes

• This is a live spike test, not a CI-safe example.
• Real Discord and model-provider credentials are required.
• Cleanup is automatic on normal completion and on Ctrl-C.
• claw-pod.yml is a spike template for the sequential test flow, not a directly runnable concurrent shared-identity topology pod.