We owe users an apology for the v2.1.0 Cognitive Sandwich installer behavior.
The initial v2.1.0 installer made the new hook stack feel mandatory, including the local Sanhedrin / MLX path. That was wrong for users on x86, Linux, Intel Mac, and any machine that does not have enough RAM for a 19 GB local model. During the hotfix, we also caught another issue: the preflight layer included preflight-swarm.sh, which can call claude -p, so preflight also cannot be enabled by default.
The corrected v2.1.0 hotfix is now published at tag v2.1.0 (1d95347) and release notes have been updated.
Default install behavior now:
- activates zero Vestige Claude Code hooks
- makes zero automatic model calls
- does not wire preflight hooks
- does not wire Sanhedrin / Stop hooks
- does not start MLX
- does not require Apple Silicon
- does not require a 19 GB model download or 20+ GB RAM
- removes old v2.1.0 Vestige hook wiring from
~/.claude/settings.json
- removes the old
com.vestige.mlx-server launchd job on reinstall
- preserves unrelated user hook entries
Cleanup command for existing v2.1.0 users:
curl -fsSL https://raw.githubusercontent.com/samvallad33/vestige/v2.1.0/scripts/install-sandwich.sh | bash -s -- --force
Optional layers are explicit:
# Optional preflight layer. This includes preflight-swarm.sh, which can call Claude Haiku.
./scripts/install-sandwich.sh --enable-preflight
# Optional Sanhedrin with any OpenAI-compatible endpoint, including x86/Linux/Intel Mac.
./scripts/install-sandwich.sh --enable-sanhedrin \
--sanhedrin-endpoint=http://127.0.0.1:11434/v1/chat/completions \
--sanhedrin-model=qwen2.5:14b
# Optional Apple Silicon local MLX autostart only when explicitly requested.
./scripts/install-sandwich.sh --enable-sanhedrin --with-launchd
Validation completed:
cargo test --workspace --no-fail-fastcargo build --release --workspace- shell/Python/JSON validation for hook scripts/fragments
- installer migration tests for default cleanup, custom hook preservation, preflight opt-in, Sanhedrin opt-in, and combined opt-in
- local reinstall verified no v2.1.0 Cognitive Sandwich hooks are wired and no MLX server is running
Thank you to @Bodge-IT and @Jelloeater for pushing on this. The default contract is now conservative: no hooks, no model calls, no local MLX, and no platform/RAM requirement unless the user explicitly opts in.
We owe users an apology for the v2.1.0 Cognitive Sandwich installer behavior.
The initial v2.1.0 installer made the new hook stack feel mandatory, including the local Sanhedrin / MLX path. That was wrong for users on x86, Linux, Intel Mac, and any machine that does not have enough RAM for a 19 GB local model. During the hotfix, we also caught another issue: the preflight layer included
preflight-swarm.sh, which can callclaude -p, so preflight also cannot be enabled by default.The corrected v2.1.0 hotfix is now published at tag
v2.1.0(1d95347) and release notes have been updated.Default install behavior now:
~/.claude/settings.jsoncom.vestige.mlx-serverlaunchd job on reinstallCleanup command for existing v2.1.0 users:
curl -fsSL https://raw.githubusercontent.com/samvallad33/vestige/v2.1.0/scripts/install-sandwich.sh | bash -s -- --forceOptional layers are explicit:
Validation completed:
cargo test --workspace --no-fail-fastcargo build --release --workspaceThank you to @Bodge-IT and @Jelloeater for pushing on this. The default contract is now conservative: no hooks, no model calls, no local MLX, and no platform/RAM requirement unless the user explicitly opts in.