coding-agents-skills
Reusable skills for coding agents
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/ggarber--coding-agents-skills)HTML
<a href="/listings/ggarber--coding-agents-skills"><img src="/api/listings/ggarber--coding-agents-skills/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
From GitHub, as a Claude Code plugin (recommended)
The repo is a Claude Code plugin marketplace, so Claude Code can install it straight from GitHub. There's nothing to clone or copy. In a Claude Code session:
/plugin install coding-agents-skills --marketplace ggarber/coding-agents-skills
Or from a shell (pick the scope):
claude plugin marketplace add ggarber/coding-agents-skills claude plugin install coding-agents-skills@coding-agents-skills --scope user # the whole computer claude plugin install coding-agents-skills@coding-agents-skills --scope project # this project, committed in .claude/settings.json claude plugin install coding-agents-skills@coding-agents-skills --scope local # this project, just you
Plugin skills are namespaced: run `/coding-agents-skills:issue-monitor`, or just ask ("monitor the GitHub issues").
- **Updates:** `claude plugin marketplace update coding-agents-skills && claude plugin update coding-agents-skills@coding-agents-skills`, or turn on auto-update for the marketplace in `/plugin` → Marketplaces. The plugin has no pinned `version`, so every new commit is an update. - **Teammates (project scope):** the committed `.claude/settings.json` declares the marketplace and enables the plugin. Each teammate still runs the `claude plugin install ... --scope project` line once per clone. - **Remove:** `claude plugin uninstall coding-agents-skills@coding-agents-skills`.
Without the plugin system
Claude Code also loads skills from plain folders:
| Where | Path | Who gets it | | --- | --- | --- | | The whole computer | `~/.claude/skills/<name>/` | You, in every project | | One project | `<project>/.claude/skills/<name>/` | Everyone who works on that project |
A skill is just its folder (`SKILL.md` plus scripts), so installing means putting that folder, or a link to it, in one of those paths. Start a new Claude Code session afterwards. The skill shows up in `/skills`, and you can call it by name (`/issue-monitor`) or just ask for it.
On the whole computer
Clone once and link the skills, so a `git pull` updates them everywhere:
git clone https://github.com/ggarber/coding-agents-skills.git ~/projects/coding-agents-skills mkdir -p ~/.claude/skills # one skill ln -s ~/projects/coding-agents-skills/skills/issue-monitor ~/.claude/skills/issue-monitor # or every skill in the repo (re-run after pulling new skills) for d in ~/projects/coding-agents-skills/skills/*/; do ln -sfn "$d" ~/.claude/skills/"$(basename "$d")"; done
Update: `git -C ~/projects/coding-agents-skills pull`.
In one project, shared through its repo
Add this repo as a git submodule and link the skill into `.claude/skills`. The link and the submodule are committed, so anyone who clones the project gets the skill at the version you pinned:
cd <project> git submodule add https://github.com/ggarber/coding-agents-skills.git .claude/vendor/coding-agents-skills mkdir -p .claude/skills ln -s ../vendor/coding-agents-skills/skills/issue-monitor .claude/skills/issue-monitor git add .gitmodules .claude && git commit -m "chore: add issue-monitor skill"
- Cloning the project: `git clone --recurse-submodules <url>`, or `git submodule update --init` in an existing clone. Without it the link points at an empty folder and the skill doesn't load. - Updating to the latest skills: `git submodule update --remote .claude/vendor/coding-agents-skills`, then commit the new submodule commit. - Add the project's own rules for the skill in `.claude/<name>.md` (see below).
In one project, as a plain copy
No submodule, no link: copy the folder in and commit it. It won't update by itself; copy it again to update.
mkdir -p <project>/.claude/skills cp -R ~/projects/coding-agents-skills/skills/issue-monitor <project>/.claude/skills/
Check and remove
ls -l ~/.claude/skills/ <project>/.claude/skills/ # what's installed and where links point
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 12 lines
- {
- "name": "coding-agents-skills",
- "description": "Reusable skills for coding agents",
- "owner": { "name": "Gustavo Garcia" },
- "plugins": [
- {
- "name": "coding-agents-skills",
- "source": "./",
- "description": "Reusable skills for coding agents: issue-monitor (GitHub issues to merged PRs, with webhook push)"
- }
- ]
- }
README.md · README · 183 lines
- # coding-agents-skills
- Reusable skills for coding agents (Claude Code and others that read `SKILL.md` skills). Each skill lives in `skills/<name>/` with a `SKILL.md` and any scripts it needs, and stays project-agnostic: project specifics go in the project, not here.
- | Skill | What it does |
- | --- | --- |
- | [`issue-monitor`](skills/issue-monitor/SKILL.md) | Watches a GitHub repo, turns each new issue into a tested, merged PR, and keeps answering comments. Events are pushed to the machine through GitHub webhooks, with cheap polling as a fallback. [`multi.py`](#watching-several-projects-multipy) runs it on several projects at once. |
- ## Install
- ### From GitHub, as a Claude Code plugin (recommended)
- The repo is a Claude Code plugin marketplace, so Claude Code can install it straight from GitHub. There's nothing to clone or copy. In a Claude Code session:
- ```
- /plugin install coding-agents-skills --marketplace ggarber/coding-agents-skills
- ```
- Or from a shell (pick the scope):
- ```bash
- claude plugin marketplace add ggarber/coding-agents-skills
- claude plugin install coding-agents-skills@coding-agents-skills --scope user # the whole computer
- claude plugin install coding-agents-skills@coding-agents-skills --scope project # this project, committed in .claude/settings.json
- claude plugin install coding-agents-skills@coding-agents-skills --scope local # this project, just you
- ```
- Plugin skills are namespaced: run `/coding-agents-skills:issue-monitor`, or just ask ("monitor the GitHub issues").
- - **Updates:** `claude plugin marketplace update coding-agents-skills && claude plugin update coding-agents-skills@coding-agents-skills`, or turn on auto-update for the marketplace in `/plugin` → Marketplaces. The plugin has no pinned `version`, so every new commit is an update.
- - **Teammates (project scope):** the committed `.claude/settings.json` declares the marketplace and enables the plugin. Each teammate still runs the `claude plugin install ... --scope project` line once per clone.
- - **Remove:** `claude plugin uninstall coding-agents-skills@coding-agents-skills`.
- ### Without the plugin system
- Claude Code also loads skills from plain folders:
- | Where | Path | Who gets it |
- | --- | --- | --- |
- | The whole computer | `~/.claude/skills/<name>/` | You, in every project |
- | One project | `<project>/.claude/skills/<name>/` | Everyone who works on that project |
- A skill is just its folder (`SKILL.md` plus scripts), so installing means putting that folder, or a link to it, in one of those paths. Start a new Claude Code session afterwards. The skill shows up in `/skills`, and you can call it by name (`/issue-monitor`) or just ask for it.
- #### On the whole computer
- Clone once and link the skills, so a `git pull` updates them everywhere:
- ```bash
- git clone https://github.com/ggarber/coding-agents-skills.git ~/projects/coding-agents-skills
- mkdir -p ~/.claude/skills
- # one skill
- ln -s ~/projects/coding-agents-skills/skills/issue-monitor ~/.claude/skills/issue-monitor
- # or every skill in the repo (re-run after pulling new skills)
- for d in ~/projects/coding-agents-skills/skills/*/; do ln -sfn "$d" ~/.claude/skills/"$(basename "$d")"; done
- ```
- Update: `git -C ~/projects/coding-agents-skills pull`.
- #### In one project, shared through its repo
- Add this repo as a git submodule and link the skill into `.claude/skills`. The link and the submodule are committed, so anyone who clones the project gets the skill at the version you pinned:
- ```bash
- cd <project>
- git submodule add https://github.com/ggarber/coding-agents-skills.git .claude/vendor/coding-agents-skills
- mkdir -p .claude/skills
- ln -s ../vendor/coding-agents-skills/skills/issue-monitor .claude/skills/issue-monitor
- git add .gitmodules .claude && git commit -m "chore: add issue-monitor skill"
- ```
- - Cloning the project: `git clone --recurse-submodules <url>`, or `git submodule update --init` in an existing clone. Without it the link points at an empty folder and the skill doesn't load.
- - Updating to the latest skills: `git submodule update --remote .claude/vendor/coding-agents-skills`, then commit the new submodule commit.
- - Add the project's own rules for the skill in `.claude/<name>.md` (see below).
- #### In one project, as a plain copy
- No submodule, no link: copy the folder in and commit it. It won't update by itself; copy it again to update.
- ```bash
- mkdir -p <project>/.claude/skills
- cp -R ~/projects/coding-agents-skills/skills/issue-monitor <project>/.claude/skills/
- ```
- #### Check and remove
- ```bash
- ls -l ~/.claude/skills/ <project>/.claude/skills/ # what's installed and where links point
- python3 ~/.claude/skills/issue-monitor/watch.py --help # the skill's scripts run from there
- rm ~/.claude/skills/issue-monitor # remove a linked skill (the clone stays)
- ```
- Install a skill in one place only. With the same skill both on the computer and in a project, or both as a plugin and as a folder, it's unclear which copy the agent uses.
- ### Other agents
- Agents that don't load `SKILL.md` skills can still follow them: point them at `skills/<name>/SKILL.md` from the project's `AGENTS.md` (or equivalent), for example "To work GitHub issues, follow `.claude/vendor/coding-agents-skills/skills/issue-monitor/SKILL.md`."
- ## Project-specific rules
- Skills read the project's `CLAUDE.md` / `AGENTS.md`. A skill can also read its own project file: `issue-monitor` reads `.claude/issue-monitor.md` if it exists. Put things there like which app vague issues usually mean, the checks to run, end-to-end recipes, and what must never be deployed without an explicit ask.
- ## issue-monitor
- ### Requirements
- - [`gh`](https://cli.github.com/) logged in (`gh auth status`) with write access to the repo.
- - Python 3.8+ (only the standard library).
- ### Push notifications instead of polling
- GitHub can't reach a laptop directly, but the official [`gh webhook`](https://docs.github.com/en/webhooks/testing-and-troubleshooting-webhooks/using-the-github-cli-to-forward-webhooks-for-testing) extension relays a repo's webhooks over an outbound connection. No public URL, tunnel or server is needed. When it's installed, the watcher uses it and reacts to new issues, comments, review comments and pushes within a second or two:
- ```bash
- gh extension install cli/gh-webhook
- ```
- The skill installs it itself when the agent is allowed to.
- Notes:
- - The forwarder creates a temporary webhook on the repo, so you need admin rights on it. The default `gh auth login` token is enough for your own repos; if creating the hook fails with 403/404, run `gh auth refresh -s admin:repo_hook`.
- - GitHub allows one active forwarder per repo, so run one watcher per repo at a time.
- - Stop the watcher with Ctrl-C / SIGTERM so it removes the hook. After a hard kill, the next start logs `Hook already exists` and retries until GitHub drops the old one.
- - The forwarder is meant for development. If it drops, the watcher keeps polling and retries the forward.
- - Measured: a new issue or comment reaches the watcher in about 2 s, with the poll interval set to 5 minutes.
- Without the extension, the watcher polls every 15 s with conditional requests (`If-None-Match` + ETag). An unchanged answer is a `304 Not Modified`, which [doesn't count against the rate limit](https://docs.github.com/en/rest/using-the-rest-api/best-practices-for-using-the-rest-api#use-conditional-requests-if-appropriate), so even fast polling costs almost nothing. Webhook deliveries are reported straight from their payload, and the poll catches anything they missed. Both paths share one record of what was reported, so each event is printed once.
- ### Running the watcher by hand
- ```bash
- python3 skills/issue-monitor/watch.py # repo of the current directory
- python3 skills/issue-monitor/watch.py owner/repo --branches main,dev --interval 10
- python3 skills/issue-monitor/watch.py --once --no-react # one check, no 👀 reactions
- python3 skills/issue-monitor/watch.py --help
- ```
- It prints one line per event on stdout (`NEW_ISSUE`, `NEW_COMMENT`, `NEW_REVIEW_COMMENT`, `NEW_COMMIT`) and status on stderr. In Claude Code, the skill runs it under the Monitor tool, so each line reaches the agent as soon as it's printed. State is in `~/.cache/coding-agents-skills/watch/<owner>_<repo>.json`. Delete it to replay the last day (commits seen on the first run are recorded, not reported).
- Comments containing `<!-- claude-issue-monitor -->` are skipped. The agent ends all of its comments with that marker, so it never answers itself when it posts from the owner's account. Change it with `--marker`.
- ## Watching several projects: multi.py
- `multi.py` starts one Claude Code session per project folder, each running `issue-monitor`. For each folder it first copies the skill into `<folder>/.claude/skills/issue-monitor`, then starts the session with `/issue-monitor`. Python 3.8+, standard library only (PyYAML is used if present).
- ```bash
- cp config.example.yaml config.yaml # then list your folders in it
- ./multi.py # folders from config.yaml
- ./multi.py ~/projects/app-one ~/projects/app-two # or pass them directly
- ./multi.py --dry-run # print what would happen, change nothing
- ./multi.py --copy-only # copy the skill into the folders, don't launch
- ./multi.py --stop # stop the tmux sessions
- ```
- `config.yaml` (ignored by git):
- ```yaml
- folders:
- - ~/projects/app-one
- - ~/projects/app-two
- permission_mode: auto # unattended sessions need a mode that doesn't stop at prompts
- # claude_args: [--model, opus] # extra arguments for every session
- # prompt: /issue-monitor # first message (the default)
- ```
- - **The copy:** every run refreshes it from this clone, so `git pull` here and re-run to update the projects. The copy is listed in the project's `.git/info/exclude`, so it never shows up in `git status` or gets committed. A project that already has the skill as a symlink or a committed folder keeps its own (it's skipped).
- - Each session runs in its own tmux session (`tmux attach -t issues-<folder>`) when tmux is installed, otherwise in a new Terminal window (macOS).
- - Folders that aren't the root of a git repository are skipped. Run one session per GitHub repo: GitHub allows one webhook forwarder per repo.
- - Without `permission_mode`, each session waits for you to approve its commands. Pick the mode you're comfortable leaving unattended. The skill still never deploys or merges red, but the agent pushes branches and merges PRs on its own.
- - Closing a window or `--stop` ends the session, and the watcher removes its temporary webhook on the way out.
- ## Adding a skill
- 1. `skills/<name>/SKILL.md` with `name` and `description` front matter. The description says when to use the skill. Refer to bundled scripts as `${CLAUDE_SKILL_DIR}/<script>`, which works for plugin and folder installs alike.
- 2. Keep it generic: read project rules from the project (`CLAUDE.md`, `AGENTS.md`, `.claude/<name>.md`) instead of hard-coding them.
- 3. Scripts next to `SKILL.md`, standard library only where possible, with `--help`.
- 4. Add a row to the table above, and check the plugin with `claude plugin validate .`.
- ## License
- MIT
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.