Small, MIT-licensed tools we built for our own work and published as they proved out. The tools are open; the ARCS intelligence that orchestrates them stays private. Nothing here claims an install path that is not live.
/plugin marketplace add flashesofbrilliance/arcs-skillsFive skills in one marketplace: rca, state-sync, token-savings-bank, fix-toolbar-comments, resume-tailor. Or reach for a single tool below. CLI tools also ship on Homebrew (brew tap flashesofbrilliance/tap) and npm.
Five steps between an idea and a message, each stamped go or stop by rules you can read. It checks the facts you mark, not whether they are true. Runs in your browser; nothing is uploaded.
Tailor a resume and cover letter to a role, graded by how checkable every line is.
Write one script; run it byte-identical across bash 3.2, zsh, POSIX sh, and ksh.
A Claude Code skill that turns "something broke" into disciplined root-cause analysis: triage the effort, reproduce the real error, fix the cause not the symptom, verify by driving the failure path, and capture the root cause.
A Claude Code skill that commits a snapshot of ground-truth project state, so any session restores it from a cheap file read instead of re-deriving it.
A Claude Code skill that installs a budget-mode operating discipline for coding agents: right model tier, batch not drip, small work inline, checkpoint state.
A Claude Code skill that clears your Vercel Toolbar feedback queue end to end: fetch every open thread, implement the fix, build, commit, deploy, and resolve each one.
An in-browser mock of the ARCS scan flow, for trying the experience. It is a UX prototype: it does not parse PDFs or call a model yet.
Stdlib Python CLI that parses your LinkedIn Connections export into four honest views of your network.
Private SAT pipeline: your logic rules go to an LLM as random tokens, a second call verifies, and only you hold the decode map. The model never sees your variable names.
Real stories from building ARCS in the open, each with the receipts.
The stories ARCS is built on, each filed with a verdict and the test that would retire it.
Dated public predictions, each with its own falsifier, re-scored against the primary documents.
We are weighing an ARCS connector for Claude, Cursor, and other agents: ask about anything in the field notes and get an answer that cites its source, not a guess. It does not exist yet. This is a live demand probe, so if you would use it, say so and we will build it.
Noted, thank you. Enough signal and we ship the connector. In the meantime the tools above are open at github.com/flashesofbrilliance/arcs-skills.
These open tools feed the ARCS compass: the private intelligence that decides when, and for whom, a thing is ready to fire.
see the compass →