github-rep 0.1.0

Analyze a GitHub profile and get honest, actionable advice for building real reputation

GitHub Reputation Engine 🛠️

A self-improving "second brain" + operating system for building genuine GitHub reputation through high-quality open-source contributions — run by an AI agent (Hermes Agent or Claude Code), autonomously, one project at a time.

Owner: Basil Al Shukaili (@basilalshukaili)

Quality of contribution > volume. One merged, appreciated PR beats fifty rejected typo fixes.

What this repo is

This repo is the portable brain of the operation. It lets any agent, on any laptop, continue the same mission with full context. It contains the plan, live state, reusable playbooks, a journal of work done, lessons learned, and nightly strategic "dreams."

Start here → CLAUDE.md (operating instructions, auto-loaded by Claude Code/Hermes).

Structure

.
├── CLAUDE.md            # Operating instructions — read first (works with Claude Code, Hermes, Codex…)
├── 00-System/           # roadmap, architecture, guardrails (the constitution)
├── 01-Targets/          # the ONE active project + next action
├── 02-Repos/            # per-repo dossiers (stack, conventions, maintainers, opportunities)
├── 03-Journal/          # daily log of what was attempted/done (append-only per day)
├── 04-Playbooks/        # portable, agent-agnostic procedures (triage, PR craft, dreaming)
├── 05-Lessons/          # mistakes & wins → rules for next time
├── 06-Dreams/           # nightly reflective synthesis: patterns, connections, next moves
└── 99-Inbox/            # scratch captures

How it works

Brain (top model)  — plans, reviews every diff, writes all maintainer-facing prose, commits
   │ delegates mechanical, fully-specified work
Worker (mid model) — implements to spec, runs tests, triage scans
   │
This repo (state)  — feeds full context back to the Brain on every session
   │
Playbooks (procedure) + Journal/Lessons/Dreams (self-improvement loop)

Token discipline without quality loss: route cheap/mechanical work to cheaper models; keep all judgment and human-facing writing on the top model. No output-degrading "compression" tricks.

Multi-machine usage

• Laptop A (Hermes Agent): runs the 24/7 cadence (cron missions + nightly dreaming), reports to Telegram. Skills live in Hermes; this repo holds the portable state + playbooks.
• Laptop B (Claude Code): clone this repo, open it, and Claude Code auto-reads CLAUDE.md. The playbooks in 04-Playbooks/ are agent-agnostic, so the workflow is identical.

Pull before you start, commit + push when you finish, so both machines stay in sync.

Safety

• No secrets are committed (API keys / bot tokens live in each machine's local env).
• Agents fork → branch → PR; never push to an upstream default branch.
• Respects each project's CONTRIBUTING rules and any anti-AI-PR policies.

License

MIT — see LICENSE.