Agents Standards & Skills
Reuse coding-agent standards, writing guidance, and task workflows across supported assistants.
- Source repo
- scarletkc/agents
- Stars
- ★ 222
- Last updated
- 5d ago
- License
- Apache-2.0
- Primary language
- Python
- FA score
- 50/100 · Major gaps
At a glance
- Works with
- Universal · cross-platformCodex · Claude Code
- You'll need
- Typical use
- A developer using both Codex CLI and Claude Code wants the assistants to follow the same baseline collaboration and coding rules.
- Main limitation
- It is not an autonomous agent or hosted service; execution depends on an external coding assistant or CLI that reads the files.
- Source review
- 50/100 · Major gaps 2 safety controls not found
What does this agent do, and when should you use it?
This repository is a collection of shared standards and reusable skills for AI coding assistants, not a standalone agent application. Its main components are a project-level `AGENTS.md`, a separate `user/AGENTS.md` for personal collaboration preferences, and a `skills/` tree whose packages use the `SKILL.md` format. The listed skills cover planning, change scoping, product and UX writing, marketing copy, worktree-based PR preparation, and delegation to Codex CLI, Antigravity CLI, or Grok Build. Users can copy the base guidelines into a repository or install selected skills through GitHub CLI or the skills CLI for assistants such as Codex CLI and Claude Code. Its deliverables are Markdown instructions and supporting skill resources; a Python packaging script can also produce upload-ready ZIP archives. It fits developers and teams seeking consistent behavior across coding assistants, but it does not provide its own hosted service, model API, or agent execution engine.
The repository supplies project-agnostic rules in AGENTS.md, keeps user-level collaboration preferences in user/AGENTS.md, and defines reusable packages under skills/<skill>/SKILL.md. A compatible coding assistant reads those files and follows the selected workflow for guided planning, documentation or copywriting, scoped changes, worktree PR preparation, or delegation to Codex CLI, Antigravity CLI, and Grok Build. gh skill install scarletkc/agents and npx skills add scarletkc/agents browse and install skills, with supported options for selecting one skill and a target assistant. python scripts/package_skill.py <skill> writes dist/<skill>.zip with SKILL.md and its supporting files at the archive root while excluding root-level evals/ and local build artifacts. After edits to the shared task runtime, python scripts/sync_agent_runtime.py updates the bundled runtime copies carried by the CLI skills.
- A developer using both Codex CLI and Claude Code wants the assistants to follow the same baseline collaboration and coding rules.
- A team maintaining several repositories wants to reuse a base
AGENTS.mdwhile prepending structure, commands, and domain notes for each project. - A user wants one bounded workflow, such as
scoped-changefor limiting an edit orask-to-planfor turning a rough idea into an actionable plan. - A product or technical writer needs to choose between the compact
product-writingguide and the more prescriptiveux-writingrules. - A skill maintainer needs an upload-ready ZIP containing
SKILL.mdand its supporting resources for a web application. - A developer already working with Codex CLI, Antigravity CLI, or Grok Build wants a dedicated delegation workflow for that command-line product.
How do you install or deploy this agent?
You need a shell, network access, and permission to write to the destination skill directory. No model API credential is documented as a repository requirement. To browse interactively and install for the current Codex user with GitHub CLI, run gh skill install scarletkc/agents --agent codex --scope user. For a non-interactive single-skill installation, run gh skill install scarletkc/agents <skill> --agent codex --scope user. In an environment that supports npx, the alternative interactive global command is npx skills add scarletkc/agents -g; to choose a skill and assistant non-interactively, run npx skills add scarletkc/agents --skill <skill> --agent codex -g -y. Replace codex with another supported assistant, such as claude-code, when appropriate.
How do you use this agent?
Review the available skills and select only the workflows you need; each skill's full activation description and instructions live in its own skills/<skill>/SKILL.md. For repository-level guidance, copy the root AGENTS.md into the target repository and prepend that project's structure, commands, and domain notes. Keep personal collaboration preferences separate using user/AGENTS.md. Once installed, give the chosen coding assistant a task matching the skill's activation description so it can read and apply the relevant SKILL.md. Update all GitHub CLI-installed skills with gh skill update --all, or update global skills CLI installations with npx skills update -g -y. To create an uploadable archive, run python scripts/package_skill.py <skill> and use the resulting dist/<skill>.zip.
What are this agent's strengths and limitations?
- It separates project guidance, user collaboration preferences, and task-specific skills into distinct reusable scopes.
- It explicitly targets Codex CLI and Claude Code while using file formats that other readers of
AGENTS.mdorSKILL.mdcan consume. - The catalog covers planning, scope control, writing, promotional content, worktree PR preparation, and multiple CLI delegation workflows rather than one generic prompt.
- It provides two documented installation paths with individual skill selection, plus a script for producing standalone ZIP packages.
- CLI skills bundle their shared task runtime, and the repository includes a synchronization step whose consistency is checked by the test suite.
- It is not an autonomous agent or hosted service; execution depends on an external coding assistant or CLI that reads the files.
- The source does not provide a skill-by-skill compatibility matrix, so behavior outside Codex CLI and Claude Code requires verification.
- Installation depends on GitHub CLI or an
npx-capable skills CLI and requires network and local filesystem access. - Packaging and runtime synchronization use Python scripts, but no required Python version or complete environment specification is given.
- The README exposes only skill summaries, so adopters must inspect each
SKILL.mdbefore accepting its triggers and detailed instructions.
How does this agent compare with similar options?
product-writing is presented as the compact option for product copy and technical documentation, leaving more structural and editorial decisions to the task. ux-writing retains a more detailed rule set for clarity, consistency, durable facts, and removal of intermediate state. The repository recommends choosing one of the two guides for a task.
Key facts side by side with the most closely related agents.
| Agent | Source review | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|
| Agents Standards & Skills This agent | 50 · Major gaps | ★ 222 | 5d ago | Python | Codex · Claude Code |
| oh-my-agent | 48 · Major gaps | ★ 1.3k | today | TypeScript | Codex · Claude Code |
| Self-Learning Skills | 56 · Major gaps | ★ 959 | 9d ago | — | Codex · Claude Code |
| Finding-Unknowns Skills | 76 · Good | ★ 338 | 7d ago | Python | Claude Code · Claude.ai |
How does FollowAgents rate this agent?
Why each dimension lost points
CI explicitly limits permissions to contents: read, installation supports selecting individual skills, and the fixtures cover permission denial and authentication failure. However, the supplied files do not systematically document per-skill data flows, sensitive-data handling, network access, or effective command permissions. External CLIs, npm execution, and GitHub Actions use version tags rather than immutable commits, leaving supply-chain controls thin. The worktree skill suggests some isolation and recovery value, but no general rollback procedure is shown. Apache-2.0, repository ownership, and the NOTICE path provide useful attribution, although NOTICE itself is absent and publisher identity remains unknown.
The README's catalog, installation commands, and packaging instructions are internally coherent, while the synchronization test checks that independently installed skills carry the canonical runtime. Offline fixtures model startup hangs, malformed JSON, process exits, timeouts, permission rejection, session mismatch, and child-process behavior with concrete diagnostics, providing meaningful failure-message evidence. The actual runtime and main behavioral tests are not supplied, so correct handling of those cases cannot be confirmed. Availability and version requirements for gh, npx, Python, Claude Code, Codex CLI, Grok, and Antigravity are also incompletely documented.
The repository clearly targets several coding agents and distinguishes project-level rules, user preferences, writing, planning, scoped changes, delegation, and worktree scenarios. Skill summaries establish broad capability boundaries, and the README explains the product-writing versus ux-writing choice. Full SKILL.md files are absent, however, so trigger precision, overlap, and false activation cannot be reviewed. Multiple agents and two installer paths show reasonable environment fit, but supported versions, operating-system differences, and a compatibility matrix are missing.
The README has a clear contents index, generated skill catalog, installation, packaging, contribution, and licensing structure, justifying full credit for information architecture. Installation examples cover interactive and unattended use but omit explicit prerequisites, removal, and troubleshooting. Naming is consistent and catalog integrity is checked in CI, though no stability or deprecation policy is stated. Examples are primarily commands, with no real FAQ and little consolidated limitations guidance. The complete Apache-2.0 text is present and NOTICE is referenced. No releases or changelog are evidenced; maintenance is limited to an Issues/PR route without named responsibility, support commitments, or a release process.
The skill catalog, selective installation, independent packaging, and cross-agent design make the output practically adoptable, while shared standards and reusable runtimes offer plausible value beyond scattered prompts. The detailed protocol fixtures suggest that CLI delegation skills address nontrivial process boundaries. Deductions apply because no representative end-user outputs, comparative evaluations, measured outcomes, or adoption evidence are supplied. External CLI setup, package downloads, and upkeep impose costs that are neither quantified nor balanced with guidance on when a skill is not worthwhile.
Core claims map to named repository paths, commands, and CI checks. The README catalog is generated from skill metadata and checked by the workflow, while the runtime synchronization claim has a dedicated test, giving useful cross-file corroboration. Fixtures express concrete protocol events and failure cases rather than broad assertions. Deductions apply because the full skill instructions, canonical runtime, primary test suite, NOTICE, and execution results are absent; the files establish verification mechanisms but cannot confirm all resulting behavior.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: sensitive-data handlingUse dedicated, low-privilege, revocable API keys — never production credentials — and keep secrets out of logs.
- This is a low-confidence static review; no tests, installers, or skills were executed.
- Review each relevant SKILL.md, bundled runtime, and its network, filesystem, and command permissions before installation; those key files were not supplied.
- The gh skill and npx skills paths, plus tag-based GitHub Actions, introduce external supply-chain dependencies; controlled deployments should pin and verify exact versions or commits.
- Do not treat the offline failure fixtures as proof that tests pass; no CI results or complete behavioral test implementation were supplied.
- The publisher is not verified by the enterprise registry. That means identity is unknown, not malicious, but adopters should independently verify the maintainer and NOTICE information.
FAQ
Does this repository execute code or call a model API by itself?
Does installation require an API key or paid service?
Can I install only one skill?
<skill> after the repository name, and the skills CLI supports --skill <skill>.How are project rules kept separate from personal preferences?
AGENTS.md, while user collaboration preferences are maintained separately in user/AGENTS.md.What troubleshooting guidance is available if installation fails?
npx environment, and destination write permissions; other problems can be reported through the repository's GitHub Issues.