I built a model-agnostic prompt linter that flags dead tokens and contradictions

by @hiroshisingh in c/prompt-engineering · · Resource

I built a model-agnostic prompt linter that flags dead tokens and contradictions
It checks your prompt for: known dead tokens for the model you pick (e.g. quality-salad in prose models), internal contradictions (static + fast camera), and buried key attributes. Open source, runs in-browser, no data leaves your machine.
github.com

It checks your prompt for: known dead tokens for the model you pick (e.g. quality-salad in prose models), internal contradictions (static + fast camera), and buried key attributes. Open source, runs in-browser, no data leaves your machine.

202 points · 8 comments

Comments

  • @theochen_motion801 · · 94 points

    A linter that flags CONTRADICTIONS is the feature I didn't know I needed. Half my bad gens are me asking for two incompatible things without noticing.

  • @zaranguyen_studio · · 60 points

    Contradiction detection is genuinely the highest-value check. Dead tokens waste a little; contradictions actively sabotage the gen.

  • @ethanbrown_art · · 52 points

    Does the dead-token database let the community contribute? That list is only as good as its coverage and it'll rot fast as models change.

  • @sakurasantos_films1880 · · 44 points

    Yes, it's a community-editable JSON with sources required per entry (so it's not just 'someone's vibe'). PRs very welcome.

  • @rileytanakacreative · · 38 points

    In-browser, no data leaves the machine, open source. This is how prompt tooling should be built. Starred and contributing.

  • @sakurapatellab4455 · · 27 points

    Feature req: let me paste two prompts and diff what the linter thinks changed semantically. For my one-variable tests.