v1.0.0

A bounded agent fan-out whose findings stream to a live local page a human answers from. One drop-in skill directory each for Claude Code, Codex CLI and grok CLI.

Download streaming-sprint-dashboard-2026-09-22.zip, unzip, then:

./install.sh claude     # or codex, or grok

Python 3.10 or newer, standard library only, no install step and no network access.

In this release

• sprint.py: init, serve, log, set, add, answer, replies, check, findings, demo
• One HTML page, rendered server side, that updates its regions in place and captures answers back to state/_feedback.jsonl
• A durable FINDINGS.md generated from the same state, opening with a decision table
• 180 tests for the skill and 25 for the package, including one suite that executes every shell block in every SKILL.md with the launcher stubbed

Verified

The grok variant has run for real: three live agents against the package itself, one question each, finishing through check and set --status done. The page has been clicked in a real browser. Contrast ratios and phone-width layout are asserted as numbers. The Codex launch command is proven at the shell level but codex exec has not run a sprint end to end.