agentsclimarketplace
Sign in

Gaslighter

Skill LarryGF/gaslighter/skills/gaslighter

Requirement completeness checker: configure nudges, run gaslighter evals, or judge a completed eval run.From its SKILL.md

Install
npx -y skills add LarryGF/gaslighter --skill gaslighter

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 5 stars5 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

1.7 KB, as published. Nobody here has run it

Gaslighter

Use this playbook to route gaslighter requests.

Routing

  • eval, evals, benchmark — load the eval skill and preserve remaining arguments.
  • judge, score, rate — load the judge skill and preserve remaining arguments.
  • config, configure, setup — load the config skill.
  • help — explain Modes and How It Works below.
  • No argument — use the host's interactive question tool to ask whether to run evals, judge a run, configure, or show help; then follow the matching route.

Use the host's skill-loading mechanism. Do not assume a harness-specific skill tool name.

Modes

Mode and nudge cap are persisted in gaslighter's config.json. Configure them through the config skill rather than re-exporting environment variables every session.

Precedence, independently for mode and cap: environment variable > persisted config > mode default.

  • lite (default) — soft nudge; default cap 3/session.
  • full — hard block; unlimited by default.
  • smart — asks a cheap model whether a requirement was missed; default cap 2/session.
  • off — disabled.

Per-session overrides:

  • GASLIGHTER_MODE — off, lite, full, or smart.
  • GASLIGHTER_MAX_NUDGES — positive integer, infinite, unlimited, or -1.

How It Works

At the end of a changed coding turn, gaslighter asks the agent to re-check the original request. It nudges at most once per attempted completion and respects the mode's cap. A subsequent tool-free re-check or an explicit 100%-certainty declaration ends the nudge cycle.

Keep looking

Skills are one crate of 627,801. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.