https://github.com/blendsdk/claude-codeops
CodeOps — an AI-development workflow for Claude Code: planning, requirements, audits, docs & git, as an installable plugin.
https://github.com/blendsdk/claude-codeops
ai anthropic claude claude-code codeops developer-tools documentation llm plugin skills workflow
Last synced: 1 day ago
JSON representation
CodeOps — an AI-development workflow for Claude Code: planning, requirements, audits, docs & git, as an installable plugin.
- Host: GitHub
- URL: https://github.com/blendsdk/claude-codeops
- Owner: blendsdk
- License: mit
- Created: 2026-06-22T07:21:48.000Z (3 days ago)
- Default Branch: master
- Last Pushed: 2026-06-22T11:57:05.000Z (2 days ago)
- Last Synced: 2026-06-22T13:06:50.969Z (2 days ago)
- Topics: ai, anthropic, claude, claude-code, codeops, developer-tools, documentation, llm, plugin, skills, workflow
- Language: Shell
- Homepage: https://blendsdk.github.io/claude-codeops/
- Size: 192 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGES.md
- License: LICENSE
Awesome Lists containing this project
README
# CodeOps for Claude Code
The CodeOps AI-development workflow — **9 skills + 13 slash commands + always-on coding
standards** — packaged as an installable [Claude Code plugin](https://code.claude.com/docs/en/plugins).
Ported from the original [`codeops-mcp`](https://github.com/blendsdk/codeops-mcp) server (built for
Cline) to native Claude Code. The MCP server existed to load rule documents on demand so the context
window wouldn't blow up; Claude Code does that natively via **skill progressive disclosure** (only
each skill's name + description load up front, the body loads when used). So the machinery is gone
and only the knowledge remains. See [CHANGES.md](CHANGES.md) for the full migration record.
> 📖 **Documentation site:** — full guides, a usage page
> for every skill, and end-to-end tutorials.
>
> **New here?** Follow the step-by-step [TUTORIAL.md](TUTORIAL.md) — it walks install → verify →
> use → update on a fresh machine.
## Install (recommended): the plugin
Inside Claude Code, add this repo as a marketplace and install the plugin:
```text
/plugin marketplace add blendsdk/claude-codeops
/plugin install codeops@codeops-marketplace
```
That's it — skills, commands, **and the always-on coding standards are active immediately**. No
`CLAUDE.md` editing and no manual merge: a bundled `SessionStart` hook injects the standards into
every session automatically (see [Always-on standards](#always-on-standards) below).
> The marketplace must be added **via git** (`add /`), as shown — that is what lets the
> plugin's `source: "."` resolve to the repo root. A raw-URL add will not resolve it.
## Verify it loaded
Inside Claude Code:
- Ask **"What skills are available?"** — the CodeOps skills should be listed.
- Type `/` and look for the namespaced commands `/codeops:make_plan`, `/codeops:exec_plan`,
`/codeops:gitcm`, `/codeops:gitcmp`, … (plugin skills/commands are namespaced under `codeops:`).
- Run `/doctor` to confirm no skill descriptions are being truncated.
## Updating
This plugin uses **rolling updates** — it carries no version number, so the git commit is the
version. To pick up the latest:
```text
/plugin update codeops@codeops-marketplace
```
Every push to the repo is immediately installable; there is no version bump to wait for. A fresh
session may be needed for newly added directories to be watched.
## Uninstalling
```text
/plugin uninstall codeops@codeops-marketplace
```
Disabling or uninstalling the plugin also turns off the standards hook — there is nothing left
behind in your `~/.claude/CLAUDE.md` to clean up (because the plugin never edited it).
## Always-on standards
The plugin bundles a single source of truth for the universal coding/testing/working-style
standards: [`standards/coding-standards.md`](standards/coding-standards.md). A `SessionStart`
hook ([`hooks/hooks.json`](hooks/hooks.json)) `cat`s that file into the context of every new
session, so the standards are always present with **zero setup**.
- This fires on every session start (including after `/clear` and context compaction).
- It is read-only — the hook only reads a file shipped inside the plugin; it never writes anything.
- To turn it off, disable the plugin. There is no separate toggle.
- If you prefer the classic approach, you *can* still merge `standards/coding-standards.md` into your
global `~/.claude/CLAUDE.md` by hand — but with the plugin that is unnecessary and would duplicate
the content.
## Developing the skills (optional)
If you want to live-edit the skills themselves, there is an in-repo dev installer that symlinks the
`skills/` and `commands/` folders into `~/.claude/` so a `git pull` propagates edits with no
reinstall:
```bash
./install.sh # symlink skills + commands into ~/.claude/ (recommended)
./install.sh --copy # install detached copies instead of symlinks
./install.sh --dry-run # preview, change nothing
./uninstall.sh # reverse it (restores any backups)
./scripts/validate.sh # pre-push guard: validates manifests, hook, standards, descriptions
```
The dev installer gives you **short** names (`/make_plan`); the plugin gives you **namespaced**
names (`/codeops:make_plan`). **Pick one — the plugin OR the dev installer, not both** — or you will
see duplicate skills. Note the dev installer does *not* install the standards hook (that is a plugin
mechanism); see [Always-on standards](#always-on-standards) for the manual-merge alternative.
## What's here
```text
codeops-skills/ # repo root == plugin root
├── .claude-plugin/
│ ├── marketplace.json # marketplace manifest (source: ".")
│ └── plugin.json # plugin manifest (no version → rolling updates)
├── skills/ # 9 skills → /codeops:
│ ├── make_plan/ # create a multi-document implementation plan
│ ├── exec_plan/ # execute a plan task-by-task (commit modes)
│ ├── make_requirements/ # gather/add/review requirements (RDs)
│ ├── retro_requirements/ # reverse-engineer a codebase into requirements
│ ├── grill_me/ # relentless disambiguation interview
│ ├── preflight/ # 13-dimension quality audit of a plan/requirements
│ ├── techdocs/ # VitePress architecture docs + ADRs
│ ├── roadmap/ # feature-set lifecycle tracker
│ └── upgrade_plan/ # upgrade outdated plans/requirements
├── commands/ # 13 slash commands → /codeops:
│ ├── gitcm.md / gitcmp.md # commit (and push) with a Conventional Commit message
│ ├── analyze_project.md # generate/refresh this project's CLAUDE.md
│ ├── migrate_clinerules.md # convert a legacy .clinerules/project.md → CLAUDE.md
│ └── … # + thin alias commands that delegate to a parent skill
├── hooks/hooks.json # SessionStart hook → injects the standards every session
├── standards/coding-standards.md # always-on coding/testing/working-style standards (single source)
├── scripts/validate.sh # pre-push validation guard
├── install.sh / uninstall.sh # optional in-repo dev installer (symlink loop)
├── LICENSE # MIT
├── TUTORIAL.md # end-to-end walkthrough
└── CHANGES.md # migration record
```
## Usage at a glance
| You type / say | What happens |
|---|---|
| `make a plan` / `make_plan` | Clarifying interview → Zero-Ambiguity Gate → `plans//` doc set |
| `exec_plan [--auto-commit]` | Implements the plan task-by-task, verifying and committing per mode |
| `make_requirements` / `add_requirement` / `review_requirements` | Requirements discovery / add one RD / health-check |
| `retro_requirements [--scope ]` | Reverse-engineer a codebase into a reconstruction brief |
| `grill_me [topic]` | Relentless design-disambiguation interview |
| `preflight ` | Adversarial, codebase-grounded quality audit |
| `make_techdocs` / `review_techdocs` | Create/maintain VitePress architecture docs + ADRs |
| `make_roadmap` / `update_roadmap` / … | Track a whole feature-set across its lifecycle |
| `upgrade_plan ` / `upgrade_requirements` | Bring an outdated artifact to current standards |
| `/gitcm` / `/gitcmp` | Commit (and push) with a detailed Conventional Commit message |
| `/analyze_project` | Generate/refresh this project's `CLAUDE.md` |
| `/migrate_clinerules` | Convert a legacy `.clinerules/project.md` into `CLAUDE.md` |
The consolidated skills cover several verbs each, and thin **alias commands** make each verb directly
typeable (they delegate to the parent skill in the right mode): `/add_requirement`,
`/review_requirements` → `make_requirements`; `/make_techdocs`, `/review_techdocs` → `techdocs`;
`/make_roadmap`, `/update_roadmap`, `/review_roadmap`, `/archive_roadmap` → `roadmap`;
`/upgrade_requirements` → `upgrade_plan`. These aliases are manual-only — only the parent skills
auto-trigger from natural language.
The skills compose into the original CodeOps pipelines, e.g.
`grill_me → make_requirements → preflight → make_plan → preflight → exec_plan`, with `roadmap`
tracking it all and `techdocs` keeping architecture docs current.
## License
MIT — see [LICENSE](LICENSE).