Open-Source Octopus Investigation

A systematic method for discovering relevant open-source projects and contributors by crawling social graphs on GitHub.

Run it as a skill. This method is packaged as a ready-to-run skill for Claude Code, Codex, and Cursor - it scopes the search with you, crawls the social graph via the gh CLI, organizes findings into tentacles, and keeps a dedup tracker automatically. Source: github.com/thdelmas/open-source-octopus-investigation.

What is it?

An octopus investigation works like its namesake - you start from a single origin (a person or topic) and extend tentacles outward through their network. Instead of keyword search, you follow human curation: stars, follows, and contributions.

This surfaces projects that are hard to find through traditional search because you're leveraging the taste and judgment of people already working in the space.

The process

1. Pick an origin

Choose a GitHub profile or topic area relevant to your project. The origin should be someone deeply embedded in the ecosystem you're exploring.

Example: obra for AI agents, memory systems, and voice interfaces.

2. Crawl the social graph

Explore the origin's:

• Repositories - what they build
• Starred repos - what they find interesting
• Followers / following - who they're connected to
• Contributions - where they participate

3. Extend tentacles

For each promising profile or project found, do a deep dive: read their repos, check what they've starred, who they follow. Each thematic cluster becomes a named tentacle.

Example tentacles from a single origin:

• Tentacle A: Tor/Arti + Reticulum (privacy-first networking)
• Tentacle B: Yggdrasil + anti-forensics (mesh protocols)
• Tentacle C: Local inference engines (on-device AI)

4. Log everything

Record each explored profile and project in a tracker to avoid re-crawling the same networks. A simple spreadsheet or markdown table works:

5. Create issues

For each relevant discovery, file a GitHub issue documenting:

• What it is
• Why it matters to your project
• How it could be integrated

Group related discoveries into thematic issues when appropriate.

6. Track unexplored leads

Profiles discovered but not yet used as origins get queued for future octopus searches. This is where the method compounds.

Why it works

Graph-based, not keyword-based

You're following human curation - what people star and who they follow - rather than search queries. This surfaces projects that haven't optimized for discoverability but are valued by practitioners.

Deduplication matters

The tracker prevents re-crawling the same person across different searches. Without it, you waste time revisiting the same networks.

Tentacles are thematic

Discoveries are organized by what they do, not by who led you to them. This makes the output actionable rather than a flat list of bookmarks.

It compounds

Each search generates unexplored leads that seed future searches. The more you run, the richer your network map becomes - a snowball effect.

When to use it

• Starting a new project and need to map the landscape
• Evaluating build vs. integrate decisions - finding existing solutions you didn't know existed
• Building a contributor network for an open-source project
• Technology scouting for a specific domain (mesh networking, local AI, etc.)

Tips

• Start narrow. One well-chosen origin is better than five mediocre ones.
• Stars are signal. What someone stars reveals their interests more honestly than what they tweet about.
• Don't go too deep too fast. Breadth-first exploration at each level before diving into any single tentacle.
• Time-box it. An octopus search can expand indefinitely. Set a scope (e.g., "3 levels deep, 2 hours max") before starting.
• Review tentacles periodically. Some will go cold, others will become central to your work. Prune and prioritize.