effort-gate

effort-gate

A universal EFFORT standing-order for Claude Code. It keeps the agent working the task to done-and-proven — relentlessly, proactively — and refuses the unnatural, silly, slop premature exits that only ever reach the average mean.

Why

The default pull of a trained model is to spend less: fewer tokens, fewer calls, fewer steps, an earlier wrap. Left alone, that produces mean-reverting, "good enough," safe-and-shallow work. The bar should be the one the task sets — not the one that is easy. effort-gate makes the bar the rule: follow the trails, keep working, never excuse a dodge.

How (the mechanism)

Same pattern as caveman — and even closer in nature, since both govern how the model works, not what it knows:

• SessionStart (activate.py) injects the full effort doctrine, read at runtime from skills/effort/SKILL.md (single source of truth — edit the doctrine, no code change).
• UserPromptSubmit (tracker.py) re-asserts a cheap trigger every turn, so the order stays in the model's attention even as other plugins inject competing instructions.

Governance is in-generation: the doctrine shapes the work because it is present in context as the model works. There is no Stop hook and no post-turn detection — catching a dodge after the turn is the anti-caveman (it polices spent tokens instead of shaping the disposition). Prevention, not policing.

The doctrine names the only lawful stops — done+verified, a real external blocker, or the user ordered less — so "relentless" never becomes pathological: legitimate, earned stops are fine; unnatural slop exits are not.

Use

• On by default once installed (level full). Nothing to do.
• /effort off · "stop effort" · "normal mode" → disable for the session.
• /effort on → re-enable.
• /effort-gate:effort → view the doctrine.

It governs how much work you do, not how you talk. Pair with caveman for terseness — terse words, full work.

Layout

effort-gate/
├── .claude-plugin/{plugin,marketplace}.json
├── hooks/
│   ├── hooks.json                 # SessionStart + UserPromptSubmit
│   └── scripts/
│       ├── activate.py            # SessionStart — inject full doctrine
│       ├── tracker.py             # UserPromptSubmit — cheap re-assert + /effort toggle
│       └── effort_config.py       # shared flag-file helpers (symlink-safe)
└── skills/effort/SKILL.md         # the doctrine — single source of truth