StudySprint — a study plan you can trust, built with the Strands Agents SDK
StudySprint turns a syllabus into a week-by-week study plan behind a deterministic evidence gate — built with the Strands Agents SDK for the Workshop to Ship hackathon.
Update — Oct 31, 2026: StudySprint v0.2.0 is out. Five new features, all from the Now list: a trust receipt BEFORE the plan — see exactly what your syllabus proved before a single study block exists · a Today view with a guilt-free 'I missed this' · time-budget reflow (25 or 45 minutes — the plan reshapes around your day) · daily check-in with recovered-weeks streaks · plain-language mode, receipts minus the jargon. Release: https://github.com/timerloggedout-spec/studysprint/releases/tag/v0.2.0 — new v0.2.0 demo:
The problem
First-generation college students often get a syllabus and nothing else: no family playbook for turning twelve weeks of topics and due dates into a daily plan. The failure mode is quiet — a missed week compounds, and by the midterm the plan is fiction.
The agent
StudySprint is an agent built with the Strands Agents SDK. Give it a syllabus and it produces a week-by-week study plan (a learn block and a retrieval block per topic), daily recall quizzes, and — when life happens — a re-plan that folds a missed week forward instead of pretending it went fine.
The starting agent, and what it already did
StudySprint extends our D0 mandate-practice agent, stated plainly as the D1 rules require. The starting agent is the Delphi Oracle: a deterministic evidence gate that judges a proposed action against an evidence set and returns ALLOW_RECOMMENDATION / ABSTAIN / ESCALATE, failing closed. In D0 ("Give your Agent a Tool") we gave it its one new tool,
fetch_evidence — read-only artifact retrieval that assembles provenance-, freshness- and hash-stamped evidence sets. Its receipts: asserted evidence only → ABSTAIN; a fetched pull request (103 stamped items) → ALLOW_RECOMMENDATION; a credential-mutation request → ESCALATE even with material evidence.StudySprint keeps that gate and that tool verbatim (vendored in
foundation/ ), because a study plan has the same trust problem: a plan built on a claimed syllabus is a plan built on a guess.The new tool
fetch_study_evidence adds the two artifact kinds a student actually has — a syllabus and a practice set — in the same stamped evidence shape. The agent's loop is the D0 loop, domain-shifted: judge on asserted evidence (ABSTAIN) → fetch the syllabus evidence set → judge again → only then build the plan and quiz.When the Strands SDK is present, StudySprint's tools register on a Strands
Agent directly ( build_strands_agent in agent.py ); the workshop behind this entry — "Building AI Agents with Strands SDK: From Simple Loops to Multi-Agent Swarms", completed in-window on 2026-10-30, all seven modules — is where that surface was exercised end-to-end, including deployment to AgentCore Runtime.What it does, concretely
Against the sample syllabus in the repo (
examples/syllabus.json , BIO 101, 3 weeks, 6 topics), the demo produces: ABSTAIN on the asserted-only request; then, on the fetched syllabus, ALLOW_RECOMMENDATION with a 12-session plan, a week-1 recall quiz, and a re-plan carrying week 1's 4 sessions into week 2. Every run emits a hashed receipt ( evidence/demo-run.json ) — decisions, evidence set, plan, quiz — so what the agent judged is exactly what you can inspect.Honesty lines
The demo runs locally and offline; no live deployment of StudySprint itself is claimed. The sample syllabus is synthetic. The gate's kernel is a deterministic baseline — no model ballots are represented in its receipts. Where the SDK is installed the same tools run under a Strands Agent; where it is not, the identical gate loop runs standalone, and the receipt says which.
Repo + demo
Public repo: https://github.com/timerloggedout-spec/studysprint
Demo: under 3 minutes, scripted in
demo-script.md , runnable as demo.py , receipts in evidence/demo-run.json .Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article