Update mnemon 0.1.17 → 0.2.3 + add Mnemon Integration Test

[gitricko]

Summary

Updates Mnemon from 0.1.17 to 0.2.3 and adds a CI integration test that verifies the full mnemon memory pipeline across two agents (Hermes and Claude).

Changes

Dockerfile

• MNEMON_VERSION bumped from 0.1.17 to 0.2.3

CI Workflow (docker-publish.yml)

• Added Mnemon Integration Test step after the existing Docker smoke test
• 3-layer verification with explicit pass/fail assertions

Integration Test Design

The test verifies that mnemon works end-to-end across two independent agents:

Step 1: Hermes remembers → hermes chat -q stores a unique name via mnemon_remember
Step 2: mnemon CLI confirms → mnemon recall finds the stored name (direct CLI check)
Step 3: Claude retrieves → claude -p recalls the name via mnemon (cross-agent check)

Why 3 layers?

Layer	What it verifies	Failure means
Step 1	Hermes tool call works	Hermes + mnemon_remember integration broken
Step 2	Data was actually stored	mnemon storage layer broken
Step 3	Cross-agent retrieval works	Claude + mnemon recall integration broken

If step 2 passes but step 3 fails, the issue is specifically in Claude's mnemon integration — not in storage. This pinpoints exactly where the integration breaks.

Why unique test names?

Each CI run generates a unique test ID (ci-mnemon-<timestamp>-<pid>) to prevent false passes from stale data. Previous runs cannot accidentally satisfy the grep assertions.

Why no cleanup step?

The container is ephemeral — when the CI job ends, the container is destroyed and all mnemon data with it. A cleanup step would be dead code.

Why show Claude output?

Step 3 captures and echoes Claude's full output before checking it. This makes CI logs debuggable — you can see exactly what Claude returned, not just pass/fail.

Verification

• 2 consecutive green CI runs after the test was added
• Test runs in ~10 seconds (steps 1-3 are fast CLI calls)
• No impact on existing smoke test or push-to-ghcr jobs

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.