Build a self-learning GitHub contributor workflow for OpenHuman

[Al629176]

Summary

Design and implement a self-learning OpenHuman workflow where OpenHuman uses its own GitHub account, a small cloud instance, and GitHub-backed memory to work on issues and PRs.

Problem / Context

OpenHuman should be able to operate more like an autonomous contributor: read issues, understand repository history, work on PRs, learn from prior outcomes, and improve future execution.

The intended setup is:

• OpenHuman has its own GitHub account.
• OpenHuman runs on a roughly $100/month cloud instance.
• OpenHuman uses GitHub memory/context to remember repo decisions, prior PRs, review feedback, failures, and successful patterns.
• OpenHuman can pick up issues and PR tasks in a controlled workflow.

This needs a clear architecture, permission model, safety gates, and implementation plan before it is treated as always-on automation.

Scope (optional)

Define and implement an initial end-to-end workflow:

• GitHub account authentication and repo access.
• Cloud runtime setup within the target budget.
• GitHub memory ingestion for issues, PRs, comments, reviews, CI failures, and merged outcomes.
• Task selection rules for issues/PRs.
• Work execution loop for making branches, commits, PRs, and responding to review feedback.
• Human approval gates for risky actions.
• Observability, logs, and cost controls.

Start with one repository, tinyhumansai/openhuman, before generalizing.

Acceptance criteria

• Architecture defined — Proposed workflow covers GitHub account, cloud runtime, memory, task selection, PR execution, review handling, and safety gates.
• Cloud plan documented — The $100/month instance target is documented with expected CPU/RAM/storage limits and tradeoffs.
• GitHub memory implemented — OpenHuman can ingest and retrieve relevant GitHub issues, PRs, comments, reviews, CI results, and historical decisions.
• Controlled GitHub account access — OpenHuman uses its own GitHub identity with least-privilege repo access and clear auditability.
• Issue/PR work loop implemented — OpenHuman can select an allowed task, create a branch, make changes, open/update a PR, and track review/CI state.
• Self-learning loop added — Outcomes from PRs, review feedback, failed attempts, and merged changes are stored as reusable memory for future tasks.
• Human approval gates — Risky actions such as pushing, opening PRs, modifying protected files, or responding publicly can require explicit approval.
• Cost and safety controls — Runtime has budget guardrails, rate limits, pause/stop controls, and logs for debugging.
• Regression safety — Tests or dry-run harnesses verify task selection, memory retrieval, branch/PR creation, and approval gating without touching production repos.
• Diff coverage ≥ 80% — the implementing PR meets the changed-lines coverage gate (Vitest + cargo-llvm-cov, enforced by .github/workflows/coverage.yml) when code changes are involved.

Related

• GitHub automation / self-learning contributor workflow.