Agent skillsgithub.com/trigga6006

Tools for the thing that builds your tools.

Skills are packaged workflows you drop into Claude Code or Codex — a slash command, a set of instructions, and the scripts behind them. These are the ones we build in the lab and use every day. All open source, all free to install.

03 skills · Claude Code · Codex · MIT

Flat isometric illustration of a cobalt tool-cartridge dock: three cartridges on a rail, one lifting into the housing's open bay, and two empty pedestals waiting

Fig.01The dock

Three in · room for more

The skills

quant skill artwork
01

quant

Codex · Claude Code

Turns an agent into a research desk that has to show its work. Every figure lands in an evidence ledger with its source, valuation and implied-expectations work is reproducible, and conclusions come with scenarios, proof gates, risks, and monitoring triggers rather than a verdict. Free sources only, so there is nothing to subscribe to.

  • Build a cited dossier on a ticker, sector, or macro situation
  • Pressure-test a thesis against explicit scenarios and proof gates
  • Track a watchlist with monitoring triggers instead of vibes
/quant-researchView on GitHub
co-review skill artwork
02

co-review

Claude Code

Gives Claude a colleague from another model family. It spawns persistent Codex workers with their own isolated worktrees, sends them bounded review or implementation tasks, and streams the answers back into the session — including an imagegen mode that produces the artwork on this site.

  • Get a cross-family second opinion before landing a change
  • Hand grunt work to a Codex worker while the main session keeps context
  • Generate raster artwork without leaving the session
/co-reviewView on GitHub
session-bridge skill artwork
03

session-bridge

Claude Code · Codex

Moves governed, provenance-linked context between two running sessions: locally, through a temporary room, or asynchronously through a self-hosted hub. Context packs are signed and recipient-minimized, so a handoff carries what the other side needs and nothing else.

  • Hand a task between two agents without losing the thread
  • Ask a peer session one focused question mid-task
  • Reconcile two agents' conclusions into one summary
/session-bridgeView on GitHub
04

next

More are in the bench. When a workflow we keep repeating turns out to be worth packaging, it lands here and in the repos — same shape, same slash command, same licence.

Suggest one

Installing

Drop it in and type the command.

A skill is a folder. Clone it into the directory your agent reads skills from, restart the session, and the slash command is there. Nothing to configure, no account, no key.

Claude Code

~/.claude/skills

$cd ~/.claude/skills

$git clone https://github.com/trigga6006/co-review

→ then run /co-review

Codex

~/.codex/skills

$cd ~/.codex/skills

$git clone https://github.com/trigga6006/quant

→ then run /quant-research

How we build them

01

One job each

A skill is a packaged workflow, not a framework. Each of these does one thing an agent is otherwise bad at, and stops there.

02

Free sources

Nothing here needs a paid data feed or a hosted service to be useful. If a skill can work from what you already have, it does.

03

Used in-house first

Every skill on this page is one we run daily. This site's artwork comes out of co-review; these pages get reviewed with it.

Take them

Free, and staying that way

Give your agent a bench.

Clone what is useful, fork what is close, and tell us what is missing. Every skill on this page is MIT licensed and built in the open.

/quant-research
/co-review
/session-bridge