https://github.com/ryanbbrown/global-agent-context
Simple setup to easily manage and version global coding agent files
https://github.com/ryanbbrown/global-agent-context
Last synced: 24 days ago
JSON representation
Simple setup to easily manage and version global coding agent files
- Host: GitHub
- URL: https://github.com/ryanbbrown/global-agent-context
- Owner: ryanbbrown
- Created: 2025-11-25T01:29:43.000Z (7 months ago)
- Default Branch: main
- Last Pushed: 2026-05-31T00:46:07.000Z (24 days ago)
- Last Synced: 2026-05-31T01:14:20.785Z (24 days ago)
- Language: Shell
- Homepage:
- Size: 256 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Ryan's Agent Context
Personal home base for how Ryan Brown uses coding agents across local side projects. This repo keeps global instructions, active skills, wrapped third-party skills, and repo bootstrapping scripts in one place, then links them into Claude Code and Codex.
The current workflow is intentionally small: start greenfield repos with the same agent-readable structure, let agents draft implementation plans in committed files, use generated HTML when a visual artifact is clearer than prose, and run multi-agent review when a plan or implementation needs outside pressure.
## Daily Workflow
### Start A Repo
Add this function to `~/.zshrc`:
```bash
unalias init-repo 2>/dev/null
init-repo() { /Users/ryanbrown/code/global-agent-context/scripts/init-repo.sh "$@"; }
```
From an empty project directory:
```bash
init-repo
```
The script creates a greenfield side-project repo with:
- `.plans/` for committed implementation plans, named in implementation order like `01-auth.md`, `02-billing.md`, `03-dashboard.md`
- `.reviews/` for committed multi-agent review outputs when they capture useful decision context
- `.html/` for committed generated HTML artifacts when visual explanation is useful
- `CLAUDE.md` and `AGENTS.md` project instructions
- a minimal `README.md`
- an initial commit and GitHub remote via `gh repo create`
### Plan Work
Plans are regular markdown files in `.plans/`. They are committed because they explain implementation order and intent over time. The filename number is the order the work is expected to land, not a priority score.
Most planning is free-form agent work rather than an explicit skill invocation. The generated project instructions carry the conventions that matter.
### Review Work
Use `personal:multi-review` when a plan or implementation needs read-only feedback from multiple agents. Review output goes under `.reviews/plans//` or `.reviews/implementations//`.
Use `personal:interview` when the agent should ask questions and shape a plan/spec before writing.
### Generate Visual Artifacts
Use `html-artifacts:html-artifacts` when HTML would communicate better than markdown: diagrams, timelines, comparison matrices, design prototypes, data explorers, or visual reports. Generated files should usually live in `.html/`.
## Active Skills
Personal skills currently kept active:
- `personal:interview`
- `personal:multi-review`
Wrapped third-party skills currently kept active:
- `gstack:browse`
- `html-artifacts:html-artifacts`
- `mattpocock-skills:grill-with-docs`
- `mattpocock-skills:improve-codebase-architecture`
- `vercel:vercel-react-best-practices`
Older personal workflows are preserved in `archive/personal-skills/` but are not linked into Claude or Codex.
## Layout
- `CLAUDE.md` is the global instruction file linked into Claude Code and Codex.
- `plugins/` is the active source tree for skills and plugin manifests.
- `archive/` preserves retired personal skills.
- `scripts/init-repo.sh` creates new side-project repositories.
- `scripts/update-skill-sources.sh` updates third-party submodules, regenerates wrapped skills, reruns links, and can commit/push known source updates from cron.
- `create-links.sh` links this repo into local Claude/Codex homes.
## Setup
Run from this repo:
```bash
./create-links.sh
```
That links:
- `CLAUDE.md` -> `~/.claude/CLAUDE.md`
- `CLAUDE.md` -> `~/.codex/AGENTS.md`
- `plugins/*` -> `~/.claude/plugins/*`
- `plugins/*/skills/*` -> `~/.codex/skills/*`
Codex's `.system` skills are preserved.
## Third-Party Sources
The repo tracks selected external skill sources as submodules and wraps only the skills Ryan currently wants active. The daily cron runs:
```bash
/Users/ryanbrown/code/global-agent-context/scripts/update-skill-sources.sh --push
```
This updates known submodules, regenerates `html-artifacts`, reruns `create-links.sh`, and commits only known submodule/generated source changes.