Claude Code v0.7.5 tester cohort: explicit remember → new SessionStart → isolated

[hjqcan]

I am looking for independent Claude Code users to validate the published GoodMemory 0.7.5 host integration. A maintainer-run smoke passed; it is evidence of compatibility, not independent adoption.

Maintainer-run environment

• macOS 15.6.1 arm64
• Claude Code 2.1.238
• GoodMemory 0.7.5 from the published npm package
• Node 22.14.0, Bun 1.3.14, npm 10.9.2
• isolated HOME / GOODMEMORY_HOME
• local SQLite, no embedding or assisted-extraction provider
• workspace_opt_in, writeback off

Observed result

1. setup registered the Claude SessionStart and UserPromptSubmit hooks plus the read-only MCP surface.
2. Project A was explicitly enabled with workspace id claude-project-a.
3. goodmemory remember --host claude ... --role user accepted one explicit fact (accepted: 1, rejected: 0).
4. A separate Claude SessionStart hook process for project A returned that fact in hookSpecificOutput.additionalContext.
5. Project B, enabled against the same local store with a different workspace id, returned no project-A marker.
6. goodmemory doctor claude reported hookRegistered: true, mcpRegistered: true, workspaceStatus: ok, zero warnings, and no planned repairs.

Reproduce

Prerequisites: Node.js 20 or newer, Bun 1.3.14 or newer, npm, and Claude Code.

npm install -g goodmemory
goodmemory setup --host claude --activation-mode workspace_opt_in --writeback off

cd /absolute/path/to/your/project
goodmemory enable claude --workspace-root "$PWD" --writeback off
goodmemory remember --host claude --workspace-root "$PWD" \
  --message "Remember that <one durable project fact>." --role user

Start a new Claude Code session in the same project and ask for that fact. Then run:

goodmemory status claude --workspace-root "$PWD" --json
goodmemory doctor claude --workspace-root "$PWD" --json

What to report

Please reply with:

• OS and architecture
• Claude Code, GoodMemory, Node, and Bun versions
• whether the new session recalled the explicit fact
• whether a second project remained isolated
• the exact failing command and sanitized output if anything failed

Boundaries

• This does not fix or replace Claude Code native auto-memory or MEMORY.md bugs.
• With writeback off, sessions do not automatically persist assistant output; the test write is explicit.
• Local SQLite is the default, but recalled content becomes model context and is visible to the model provider configured in Claude Code.
• Downloads, stars, scans, and maintainer-run tests are not counted as adoption. Only an independent success report or reproducible failure will update the adoption record.