Launching today

foolscap

Launching today

Delegate to your coding agents, and prove what came back

2 followers

Delegate to your coding agents, and prove what came back

2 followers

Visit website

Tell foolscap what needs doing. GPT-6 Astra plans the work and routes it to Claude Code, Codex or OpenCode running on your machine. Then foolscap reads the test output and the diffs itself, so a result is evidence, not the agent's word for it. Red tests trigger a bounded repair; passing work gets reviewed by a different agent. Local-first, bring your own key, MIT.

Free

Launch tags:Open Source•Developer Tools•Artificial Intelligence

Launch Team / Built With

OpenAI

GPT-6 Astra Challenge

What real task does your product handle with GPT-6 Astra?

Maker

Supervising other coding agents to a finish you can verify. You hand foolscap a task — "make the two failing replay cases pass, then have a second agent review the change" — and Astra runs it as a supervisor, not a writer. It dispatches Claude Code into your checkout as an async tool call: the call returns immediately, so Astra keeps planning and reading evidence while the agent works for minutes. Tests come back red; it reads the real output, sends one bounded repair, then dispatches the review to a different agent family, read-only — because an agent cannot sign off on its own work. Nothing is taken on trust. A deterministic evidence engine, not a model, decides whether a run passed, by classifying what the test command actually printed. Implementation repairs cap at two, reviews at two. A real run: implement failed 1 test, one repair, 1 passed, independent review on Codex, complete — 6 turns, $0.08. Every step stays on the record.

Report

4h ago

Maker

📌

I kept running three coding agents at once and losing track of which one actually did the thing. They all say "done". Reading five terminals to find out who was lying is slower than doing the work yourself. So foolscap doesn't ask. You tell it what needs doing, GPT-6 Astra plans it and hands the work to Claude Code, Codex or OpenCode on your machine — and then foolscap reads the test output and the diffs itself. Tests red, it sends a repair with the failure attached. Tests green, a *different* agent reviews the diff. The card only moves when the evidence says so. The Astra Challenge is what made the coordinator possible. Async tool calling means it can dispatch an agent and keep planning instead of blocking, and mid-run it's reading evidence rather than trusting a summary. A full run costs about eight cents of Astra; your agents do the rest. Two honest notes: the Warp and Antigravity launch paths are written from their docs and I haven't run them against a real install — the README says so rather than showing a tick. And it's local-first: your session history never leaves your machine. Would genuinely like to hear where it breaks. It's MIT, the whole thing installs with one line, and adding a new agent is a single file.

Report

17h ago

Forum Threads

p/foolscap

•

16h ago

An agent saying "done" is the least reliable part of the run. What do you check ?

I kept hitting the same thing: a coding agent finishes, says it's done, and the tests were either never run or still red. The summary was confident either way. So the actual work wasn't writing code it was checking up on it.

foolscap is what came out of that. You give it a task, GPT-6 Astra plans it and dispatches Claude Code, Codex or OpenCode on your machine, then a deterministic evidence engine not a model reads what the test command actually printed and decides whether the run passed. Two rules that do most of the work:

repairs are capped (two for implementation, one after review), so it can't grind forever
the review is dispatched to a different agent family, read-only, because an agent shouldn't sign off on its own work

A run from yesterday: implement failed 1 test one repair 1 passed independent review on Codex complete. Six turns, eight cents, every step on the record.

View all