Weekly · star velocity · 2026-06-15

Teaching AI Coding Agents Better Habits

This week's attention went mostly to rules and tools that shape how AI coding assistants work, from a prompt that curbs over-engineering to context compression and pre-install skill scanning.

Fastest-growing this period

Counts come from GitHub's official stargazers/history endpoint: the stars gained on each day of the window. Project briefs and the reading notes below are editorial additions.

1

DietrichGebert/ponytail : Stop your AI turning a small fix into a new framework
JavaScript
A set of instructions that asks AI coding agents to reuse existing code and question whether a new abstraction is needed. Its rules also explicitly retain validation and error handling, useful when your agent keeps making a small task unnecessarily complicated.

+871
2

elder-plinius/CL4R1T4S : Study a third-party collection of claimed AI system prompts

A third-party collection of purported system prompts and tool instructions from AI products. It can provide material for studying prompt structure, but the files are not authenticated examples of how those products currently work.

+330
3

chopratejas/headroom : Trim bulky tool output before it fills your AI's context
Python
Compresses material such as logs, tool outputs and retrieved documents before an agent sends it to a model. Useful when large, repetitive inputs consume the context window; test whether the compressed version preserves the details your task actually needs.

+288
4

Panniantong/Agent-Reach : Give your agent a way to read beyond ordinary web pages
Python
Gives command-line AI agents tools for reading and searching websites, video platforms and social sources. Its channel diagnostics help identify which connections work and which need configuration, useful when gathering material across several sites.

+253
5

mattpocock/skills : Make your AI ask the right questions before it writes code
Shell
Give your coding assistant a skill that asks detailed questions before implementation. Use it to clarify vague requirements before the agent builds the wrong thing.

+247
6

iptv-org/iptv : Find TV streams by language and open them in your player
TypeScript
Collects publicly available television streams into M3U playlists that compatible players can open. Useful for finding channels by category or language, with playback depending on the individual stream and your network rather than the playlist alone.

+245
7

obra/superpowers : Give a multi-step coding task checkpoints you can inspect
Shell
A skill collection that guides a coding assistant through requirements, planning, implementation, tests and review. Useful for longer features where you want explicit checkpoints before the agent moves from an idea to a large set of code changes.

+176
8

addyosmani/agent-skills : Give your coding agent a repeatable engineering routine
Shell
A set of coding-agent workflows covering specifications, implementation, tests and review. Each skill includes verification steps, useful when you want to give an assistant a repeatable engineering routine instead of improvising a fresh prompt for every task.

+166
9

Egonex-AI/Understand-Anything : Explore an unfamiliar codebase as an interactive map
TypeScript
A Claude Code plugin that turns files, functions and dependencies into an interactive code map with explanations. The generated map can be shared with teammates, making an unfamiliar repository easier to explore together.

+162
10

farion1231/cc-switch : Stop editing the same model settings in several tools
Rust
A desktop control panel for the model providers, MCP servers and skills used by several AI coding tools. Useful if you alternate between Claude Code, Codex and similar tools and are tired of keeping their configuration files in sync by hand.

+144
11

Leonxlnx/taste-skill : Show your AI the look you want before it builds the UI
JavaScript
A skill collection for improving the layout, typography and motion of AI-built interfaces. It also supports making visual references before coding, useful when a clear example would communicate your design intent better than another long prompt.

+129
12

NousResearch/hermes-agent : An assistant that can turn past tasks into reusable skills
Python
A personal AI agent with persistent memory, accessible from a terminal or chat apps. It can save skills from completed tasks and search earlier conversations, making it interesting for recurring work that benefits from remembering previous attempts.

+129
13

pewdiepie-archdaemon/odysseus : Bring AI chat, research and notes into one workspace
Python
A self-hosted workspace that brings AI chat, research, documents, notes and planning tools together. Useful for exploring whether keeping these activities in one application helps your daily workflow, with feature maturity checked against the tasks you need.

+125
14

NVIDIA/SkillSpector : Check an agent skill before you let it near your files
Python
Scans agent skill packages for suspicious instructions and credential or supply-chain risks before you install them. It can remember previously triaged findings, useful for reviewing updates without treating every old warning as a new discovery.

+117
15

affaan-m/ECC : Give your coding agent a plan, tests and a memory
JavaScript
A collection of coding-agent workflows for planning, tests, review and remembering useful context. If your assistant jumps straight into edits or forgets earlier decisions, this gives you reusable routines to adapt to your project.

+111
16

GoogleCloudPlatform/knowledge-catalog : Teach your AI what those database columns actually mean
Python
Stores information about datasets in ordinary Markdown files that people and AI tools can both read. It includes a way to turn BigQuery metadata into these documents, useful when an assistant needs context about what your tables and columns actually mean.

+108
17

colbymchenry/codegraph : Let your coding agent follow the call chain
TypeScript
Builds a local index of code symbols, calls and dependencies for AI assistants to query through MCP. Useful for tracing related code before making a change; try it on your own repository to see whether its answers are more helpful than ordinary file search.

+107
18

nexu-io/open-design : Keep your AI's visual work tied to one design brief
TypeScript
A desktop workspace that connects coding agents to visual tasks such as prototypes and presentations. It uses a shared design-instructions file, useful for keeping colors, typography and other preferences available across several related outputs.

+105
19

mvanhorn/last30days-skill : Find what people discussed in the last month
Python
A research skill for gathering recent discussion about a topic across web and community sources. Useful before writing about a fast-moving subject, with the collected posts serving as leads to verify rather than a complete or representative account of public opinion.

+103
20

omnigent-ai/omnigent : Let one coding model implement and another review
Python
Coordinates different coding agents through one interface, including workflows where one model implements and another reviews. Useful for experimenting with a division of work across tools while keeping tasks and review steps in one place.

+102
21

DeusData/codebase-memory-mcp : Ask your codebase who calls what
C
Builds a persistent graph of a codebase's functions, classes and call relationships for AI coding tools to query. Useful when you want to ask who calls a function or what a change might affect without repeatedly scanning the whole repository.

+100
22

microsoft/markitdown : Turn a pile of mixed documents into Markdown
Python
Converts documents such as PDFs and Office files into Markdown for text-processing tools. Useful when you want to feed a folder of mixed documents into search or an AI workflow while retaining headings, lists and other readable structure where supported.

+95
23

shadcn/improve : Have one model plan a code improvement and another execute it

A coding-agent skill that audits a repository and writes implementation plans for another model to carry out. Useful for separating careful analysis from execution, with verification steps attached so you can review whether the proposed improvement actually landed.

+95
24

multica-ai/andrej-karpathy-skills : Give your coding agent a few rules it can actually follow

A compact set of coding-agent instructions about stating assumptions, keeping changes small and defining verifiable goals. This community interpretation of Karpathy-inspired advice is useful if you want a few clear rules instead of a large workflow package.

+91
25

EpicGames/lore : Work on a game without downloading every asset first
Rust
Epic Games' version control system for projects containing both code and large assets such as game art. Its sparse workspaces fetch file data on demand, which is useful when downloading an entire asset-heavy repository would be unwieldy.

+88

What I would look at first

Reading notes updated 2026-09-07; the ranking above preserves its original reporting window.

Look down this list and a pattern jumps out: many of the week’s most-attended projects are not programs you run, but rules and workflows you hand to an AI coding assistant. The top entry, DietrichGebert/ponytail, drew 871 star events this week. It is a plain set of instructions telling the agent to check whether existing code can be reused before inventing a new abstraction, and to keep validation and error handling in place. If your assistant has a habit of turning a two-line change into a small framework, that is the scenario it targets.

It is one of several takes on the same problem. mattpocock/skills concentrates on clarifying the requirement: one of its workflows makes the agent interview you first, asking detailed questions before implementation. obra/superpowers and addyosmani/agent-skills go further, laying out a full routine from planning through tests to review, with verification steps attached. If your main pain is an agent that codes before it understands, start with the questioning skill. If your pain is a large set of changes nobody checked, the fuller routines are worth a look.

A different bottleneck shows up in chopratejas/headroom. Rather than changing how the agent behaves, it shrinks what the agent has to read, compressing logs, tool output and retrieved documents before they eat the context window. That is appealing when you feed an agent a long build log or a pile of search results. Compression needs checking because it can drop details. A sensible first test: compress a real log from your own project and check whether the facts your task needs survived.

For people who hop between Claude Code, Codex and similar tools, farion1231/cc-switch puts model providers, MCP servers and skills into one desktop panel, so you stop hand-editing several config files that should say the same thing.

Finally, a security note. As skills and rule packs become something you download like any other dependency, NVIDIA/SkillSpector scans them for suspicious instructions and credential risks before you install, and remembers what you already triaged so repeat warnings do not drown out new ones. Much of this week’s list is exactly that kind of downloaded instruction file, so scanning before install is a reasonable habit. As a first experiment, run it against whichever skill collection you just picked.

Teaching AI Coding Agents Better Habits

Fastest-growing this period

DietrichGebert/ponytail : Stop your AI turning a small fix into a new framework

elder-plinius/CL4R1T4S : Study a third-party collection of claimed AI system prompts

chopratejas/headroom : Trim bulky tool output before it fills your AI's context

Panniantong/Agent-Reach : Give your agent a way to read beyond ordinary web pages

mattpocock/skills : Make your AI ask the right questions before it writes code

iptv-org/iptv : Find TV streams by language and open them in your player

obra/superpowers : Give a multi-step coding task checkpoints you can inspect

addyosmani/agent-skills : Give your coding agent a repeatable engineering routine

Egonex-AI/Understand-Anything : Explore an unfamiliar codebase as an interactive map

farion1231/cc-switch : Stop editing the same model settings in several tools

Leonxlnx/taste-skill : Show your AI the look you want before it builds the UI

NousResearch/hermes-agent : An assistant that can turn past tasks into reusable skills

pewdiepie-archdaemon/odysseus : Bring AI chat, research and notes into one workspace

NVIDIA/SkillSpector : Check an agent skill before you let it near your files

affaan-m/ECC : Give your coding agent a plan, tests and a memory

GoogleCloudPlatform/knowledge-catalog : Teach your AI what those database columns actually mean

colbymchenry/codegraph : Let your coding agent follow the call chain

nexu-io/open-design : Keep your AI's visual work tied to one design brief

mvanhorn/last30days-skill : Find what people discussed in the last month

omnigent-ai/omnigent : Let one coding model implement and another review

DeusData/codebase-memory-mcp : Ask your codebase who calls what

microsoft/markitdown : Turn a pile of mixed documents into Markdown

shadcn/improve : Have one model plan a code improvement and another execute it

multica-ai/andrej-karpathy-skills : Give your coding agent a few rules it can actually follow

EpicGames/lore : Work on a game without downloading every asset first

What I would look at first