Your AI agent starts strong...then drifts, loses context, and skips steps. HyperWorker stops that. Six mechanisms in markdown and YAML keep long-running projects on track: Lock (no drift), Atomicity (no "mostly finished"), Dependency (no skipped steps), Memory (session survival), Precedence (rule conflicts), Verification (evidence-based completion). Point your agent at HARNESS.md and it scaffolds everything. Built in Cowork. Works with any LLM that reads files. Open source, MIT licensed.
I've been using AI agents for multi-session projects (not one-shot tasks...actual work with phases, dependencies, and handoffs between sessions). Curious what patterns other people are seeing.
No reviews yetBe the first to leave a review for HyperWorker
Maker
📌
Hey Product Hunt! I'm Spencer, and I built HyperWorker because I kept hitting the same wall running projects through AI agents.
The agent would be great for the first few tasks. Then it would forget what happened last session, edit files it was told not to touch, or mark something "done" based on nothing but its own confidence. The problem wasn't the model...it was the environment. There was nothing structurally preventing drift.
HyperWorker started as rules in a system prompt. When that stopped scaling (prompts compress, rules conflict, knowledge gets lost), I externalized everything to markdown and YAML files that the agent re-reads every session.
v4 adds Verification as the sixth mechanism...every task now produces an evidence trail proving what was checked and what happened. The ratchet principle means the project can only move forward.
The launch campaign for HyperWorker was managed by HyperWorker itself. That felt like the right test.
I'd love feedback from anyone running real multi-session projects through AI agents. What breaks for you? What patterns have you found?