A collection of agent skills for planning, coding, review, research, and UI work, with a small set of shared Codex defaults. It includes original workflows and adapted skills from the projects credited below. Browse skills/ and see SOURCES.md for upstream origins and local differences to preserve during updates.
Clone the repo, then link the skills you want into your agent’s skill directory.
For a host that reads ~/.agents/skills:
git clone https://github.com/sreejithraman/SreeStack.git
cd SreeStack
mkdir -p ~/.agents/skills
ln -s "$PWD/skills/tdd" ~/.agents/skills/tddRepeat the last command for other skills. If that destination already exists, keep it or move it aside before linking. To uninstall, remove only the symlink you created. Pull changes in this clone to update linked skills.
Read each skill before use. Some workflows need host tools for goals, agents,
or browser control; others need GitHub CLI, Antigravity CLI, Tailscale, or Apple
development tools. Available tools and invocation settings vary by host.
This collection has no single runtime that supports every skill.
add-to-snip-snap requires the snipsnap command on PATH; verify it with
command -v snipsnap. Installing or copying only the GUI app does not expose
the CLI automatically. A Snip Snap Homebrew cask release that declares the
bundled snipsnap binary does.
The files under codex/ define shared defaults. Keeping them here does not change the live Codex config or global instructions. Keep the full personal config local. It can contain machine paths, trusted projects, and integration settings.
| Source | Destination | How to apply |
|---|---|---|
| config.toml | ~/.codex/config.toml |
Merge the listed keys into the existing file and [agents] table |
| global-instructions.md | ~/.codex/AGENTS.md |
Merge these shared instructions into the existing file |
| optional-instructions.md | ~/.codex/AGENTS.md |
Merge only the sections you want to enable |
| optional.rules | ~/.codex/rules/default.rules |
Replace its path placeholder and merge only the rules you want to enable |
| junior.toml | ~/.codex/agents/junior.toml |
Copy this rank file |
| engineer.toml | ~/.codex/agents/engineer.toml |
Copy this rank file |
| senior.toml | ~/.codex/agents/senior.toml |
Copy this rank file |
| staff.toml | ~/.codex/agents/staff.toml |
Copy this rank file |
To apply these defaults, back up the local files, merge or copy only these parts, and
check that the TOML parses. Keep unrelated values and existing instructions.
If ~/.codex/AGENTS.override.md exists, Codex reads it instead of AGENTS.md;
merge these shared instructions into that active file, or deliberately retire
the override before using AGENTS.md.
If a destination role already exists, review its differences before replacing
it.
No install script or whole-config symlink is needed for this first version.
Orchestration and review use the routing reference for role selection and the host's spawn rules. Model and effort values live in config and agent files; the skills own delegation, review coverage, and rounds.
These are defaults for local tasks using the same Codex home. Project config, explicit model choices, and custom roles can override them. Defining roles makes them available; the skills request their use. Start a new task after applying the files and check the effective settings. Existing tasks may retain their selections. Keep any installed links on a stable clone, not a temporary worktree.
Sample checks after installation:
- Every nonempty review scope also runs the external model step in review-fix-loop. Native reviewer counts below do not include that step.
- A small behavior change uses one reviewer; substantial changes, multiple behaviors, shared contracts, risky logic, or work from several agents use two. Each reviewer reads the whole diff in every round.
- A complete clean review can finish after one round. Accepted code fixes trigger another full review against the original base.
- Ordinary docs use parent review. Docs that change agent behavior use one independent reviewer, with another full review after substantive fixes.
- Independent work can use agents without creating goals unless the user asks for goal-backed work. A difficult cohesive task can keep one owner.
Local research notes under docs/research/ are ignored by Git. Keep maintained
technical citations beside the guidance they support. The setup above follows
OpenAI's config precedence,
custom agent format,
and global instruction discovery.
See CONTRIBUTING.md for changes and checks, and AGENTS.md for agent instructions.
SreeStack’s original work uses the MIT License. Imported material keeps its own terms; see THIRD_PARTY_NOTICES.md. Report security issues through the private reporting channel.
Thank you to the authors and contributors who shared the skills this collection uses and adapts:
- Matt Pocock — engineering and productivity skills, including the Standards and Spec review rules.
- Emil Kowalski — animation, motion and gesture behavior, typography and materials, interface access/input, and interface-review guidance.
- Jakub Antalík’s transitions.dev — motion pattern references and tuning guidance used by
animate. - Dietrich Gebert’s Ponytail — simplicity and code review rules.
- Cursor — the Thermo Nuclear Code Quality Review.
- Anthropic’s skills — art-direction guidance used by
ui-design. - Anthropic’s Claude Plugins Community — the
eli5skill. - GitHub’s gh-stack — the stacked PR skill.
- s0xDk — web-system guidance used by
ui-design, based on Adam Wathan and Steve Schoger’s work. - React Doctor — the React diagnostics skill.
- Vercel’s agent-skills — the React Best Practices skill, originally created by Shu Ding.
- Charles Wiltgen’s Axiom haptics guide — inspiration for our locally written iOS haptics skill; no upstream prose or code copied.
- Rudrank Riyam’s App Store Connect CLI skills — usage and workflow guides for our App Store Connect skill.
- superagents-lab’s xcode27-skills — coverage input for independently written Xcode security and SwiftUI SDK 27 guidance; no upstream prose or code copied.
- Snip Snap — the add-to-snip-snap skill and its CLI workflow.
SOURCES.md records upstream paths, import revisions, and current differences to preserve. Git records how the skills changed over time.