pi-braid-skills

Skills for research papers, brand systems, illustrations, and design critique that load straight into the pi coding agent.

Packages

Package details

skill

Install pi-braid-skills from npm and Pi will load the resources declared by the package manifest.

npm repo home report

$ pi install npm:pi-braid-skills

Package: pi-braid-skills
Version: 0.2.5
Published: Nov 3, 2026
Downloads: 329/mo · 37/wk
Author: kifirkin
License: MIT
Types: skill
Size: 171.7 KB
Dependencies: 0 dependencies · 0 peers

Pi manifest JSON

{
  "skills": [
    "./skills"
  ]
}

Security note

Pi packages can execute code and influence agent behavior. Review the source before installing third-party packages.

README

pi-braid-skills

Ten skills for the pi coding agent: research papers, brand systems, visual work, writing that has to land, and a tunnel that has to come back. Install once, usually per project, and the agent opens a skill when the task matches it.

Skills are instructions, not tools. Installing this pack teaches the agent ten procedures; it changes what the agent knows how to do, not what it can do.

Install

pi install npm:pi-braid-skills

The pack is usually wanted per project: add the same spec to that project's .pi/settings.json (the declaration pi reads after you trust the project) rather than to your user settings.

Running from a checkout instead? Register the workspace path in ~/.pi/agent/settings.json:

{
  "packages": ["file:~/pi-braid/packages/skills"]
}

What you get

Skill	What it does	Use it when	Status
`adversarial-engagement`	Five probes that force a real answer out of a scripted one: incongruity, structural irony, world-building, binary-trap evasion, label contradiction	A reply keeps getting deflected or templated, and the exchange is going in circles	Usable
`cv-improvement`	Measures and improves a CV: metric formulas, a failure catalogue keyed to the form, the fix order, and honesty rules on unconfirmed numbers	You are improving or reviewing a CV or resume, a draft reads as machine-written, or the page budget is about to break	Unvalidated
`design-system`	Authors a `DESIGN.md` brand system: tokens for color, type, spacing and components, plus the consistency workflow	You are building a brand system, writing a `DESIGN.md`, or enforcing design consistency across a project	Usable
`paper-writer`	Structures and writes LaTeX papers for ML venues: IMRaD, abstract strategy, related work, figures, tables, citations, camera-ready conventions	You are drafting or revising a NeurIPS, ICML, ICLR or ACL submission, an arXiv preprint, or a workshop paper	Usable
`tunnel-control`	Diagnoses and restores a dropped tunnel client: reachability as ground truth, the platform's own status commands, and recovery through the client's own control surface	One service times out while other traffic works, or a tunnel reports connected but nothing loads	Unvalidated
`visual-creator`	Creates visuals that survive critique: design principles, color, typography, layout, brand identity, and the create, analyze, validate, improve loop	You are building a UI, designing game assets, making brand material, or producing a visual that needs coherent reasoning	Usable
`visual-validator`	Design critique with measurable, actionable fixes: sprite assets, game screenshots, UI designs, image quality	You need a visual audited, a UI reviewed, or an image quality call	Usable
`meme`	Static canvas memes with Impact captions, posted through the browser clipboard	You want a one-off static image meme	Deprecated
`socratic-irony`	Wit craft: incongruity as the engine, voice archetypes, dramatic structure, and where to stop	You are writing a roast, satire or sharp critique and need a sharper line	Deprecated
`twitter`	Posts replies on X through browser automation and checks whether the bot answered	You are working an X thread with a browser session	Deprecated

Three of the ten are marked deprecated in their own frontmatter: meme points at a separate project for animated work, socratic-irony overlaps the tone-quality skill, and twitter has no successor. They stay in the pack because their technique sections are useful reference; they are not what the pack is for.

The three status values say what a row is worth to you: Usable has been run against a real task, Unvalidated ships and loads but no run has yet shown the procedure working, and Deprecated stays for its reference sections rather than as the recommended way to do the work.

How a skill gets used

pi reads each SKILL.md frontmatter description and decides from that whether to open the file. You do not select a skill. Two consequences are worth knowing: a skill fires on its description, not on its folder name, and you can always ask for one by name ("use the paper-writer skill").

Limits

pi only. These are SKILL.md files; another agent's loader may ignore them.
No code, no tools, no settings keys. The pack is instruction text.
Three skills are deprecated and two are Unvalidated (see the table); the other five are the ones to install for.

License

MIT. See LICENSE.

For pack maintainers

Every SKILL.md opens with frontmatter: name matches its directory, description is non-empty and under pi's 1024-character limit, validated is an ISO date, and the verdict: field (DEPRECATE / IMPROVE / UNVALIDATED) records the last review outcome. A body that directs a tool family may declare it under governs:; most declare none, and the declaration is optional. The monorepo's contract test enforces those fields before a release; the tarball does not carry the test. Shared fragment files live in skills/_shared and carry no SKILL.md.

Releases are cut from the monorepo: the tarball carries skills/, this README, AGENTS.md, LICENSE and package.json.