← ClaudeAtlas

marginallisted

Use when inspecting, reviewing, promoting, demoting, or explaining MARGINAL governance in Codex, especially for repeated tool work, token-saving claims, and Earned Enforcement readiness.
SignalLayerLabs/Marginal · ★ 18 · AI & Automation · score 74
Install: claude install-skill SignalLayerLabs/Marginal
# MARGINAL Treat compute as scarce and claims as evidence-bound. The plugin starts globally in Shadow Mode; it may exercise repository-scoped Tool Enforcement only after a valid Earned Enforcement receipt and explicit promotion. ## Native control Do not require a global `marginal` executable or a pip installation. Resolve the native plugin: 1. Run `codex plugin list --json`. 2. Select the exact `pluginId` `marginal@marginal` and read its `source.path` as the plugin root. 3. On macOS/Linux, run `python3 <plugin-root>/scripts/marginal_control.py COMMAND`; on Windows, run `py -3 <plugin-root>\scripts\marginal_control.py COMMAND`. Pass `--workspace <repository>` and `--json` when inspecting repository-scoped state. The launcher uses Codex's native plugin data directory, so hook evidence and control commands share one state. It automatically replaces an older `python3` with an available Python 3.10–3.13 interpreter. If none is installed, report that exact runtime requirement and do not claim hooks are operational. ## Workflow 1. Run the native `status` command before describing the active mode. 2. Read live operation, prior evidence, and enforcement as separate facts: - `hooks_active: true` attests a live authenticated MARGINAL lifecycle service for this repository. It is the strongest available operational signal, but does not expose or prove a raw chat session identity. - `hooks_observed: true` proves lifecycle evidence exists for this repository, even