Project Prospector

Install

/plugin marketplace add 88plug/project-prospector
/plugin install project-prospector@project-prospector

Quickstart

Ask in plain language — no command to memorize:

What have I built on this laptop? Rank my projects by which ideas are most original.

You get a tiered ranking, strongest idea first, with [idea]/[LIVE]/[dormant] tags and an evidence-anchored one-liner for each entry:

Tier S — genius
  benchie [LIVE] — predicts engine latency without launching it
    (perfmodel/roofline.py), 72 commits this week.

Tier A — elegant, high-leverage
  searxng-mcp — token-efficient self-hosted search for agents; dual-IP failover.

Dormant
  old-scraper — last real commit pre-cutoff; cosmetic touch since.

What it does

Project Prospector surveys a whole machine for your own work and ranks it by the quality of the idea, not how finished it is. A half-built concept with a novel core can outrank a polished CRUD app. The name is the intent: you're prospecting a messy filesystem for the few strong ideas buried in it.

It runs a two-pass, parallel, read-only sweep, then synthesizes one ranking:

• Catalog pass — clusters the filesystem into themed groups (crypto, homelab, AI-tooling, not alphabetical) and gives each cluster its own read-only explorer agent that reads READMEs, runs git log since the cutoff, and judges non-git dirs by file mtimes.
• Blind-spot pass — a second wave of agents that each attack one place a file sweep structurally misses, so you don't confidently report "that's everything" and be wrong.
• Synthesize — de-duplicates, separates idea quality from execution state, and produces an S–D tiered ranking with evidence-anchored rationale and alternative lenses.

Why two passes

One agent reading directories top to bottom misses most of the value. The strong ideas hide in places a plain ls never reaches, so the blind-spot pass covers:

• Transcripts — ~/.claude/projects/ slugs, grepped for idea and plan language. Finds ideas discussed but never turned into a folder.
• Other agent CLIs — ~/.codex/, ~/.opencode/, ~/.config/ agent tools. Finds work done through other tools.
• Running services and history — shell history themes, crontab -l, systemctl --user timers and units, docker ps -a, long-running processes. Reveals what is actually live versus abandoned.
• Research artifacts — substantive docs in ~/Downloads, ~/Documents, ~/Desktop, and browser bookmarks and history. Finds ideas you're circling but haven't built.
• Beyond home — /opt, /srv, /mnt, /media, nested repos inside other projects, and system-wide recently-modified source. Confirms nothing hides outside the obvious tree.

How it ranks

The default axis is idea-novelty, non-obvious insight, and leverage — not lines of code, not polish. Output is structured as tiers, strongest first:

• Tier S — genius: genuinely novel core insight, high ceiling.
• Tier A — elegant, high-leverage: strong idea, clear payoff.
• Tier B — clever hacks, narrower: smart but bounded in scope.
• Tier C — solid, low novelty: useful and reliable, not inventive.
• Tier D — utility, scratch, and stubs, plus a Dormant list.

Each entry carries a tag that separates the idea from its execution state: [idea] (no codebase yet, can still rank at the top), [LIVE] (currently running), and [dormant] (untouched before the cutoff). Every cited path, commit count, or number is verified against disk before it goes in the report.