snappedly/skills

**Engineering workflow skills for AI coding agents.** Plan a change, build it, review it, and merge it, with the checks a careful engineer runs at each step.
Snappedly Skills is a collection of Agent Skills: Markdown instructions that your coding agent loads when a task calls for them. They work in Claude Code, Codex, and the other agents that the Skills CLI supports.
Each skill does one job, such as stress-testing a plan, reviewing a diff, or diagnosing an intermittent bug. The skills hand work to each other. They read your team's conventions from files in your repository, so one set of skills fits teams with different hosts, trackers, and review rules.
> [!IMPORTANT]
> These skills direct an AI agent that can edit files, run commands, push code, and merge pull requests. Read the disclaimer before you install them.
Quick start
1. Install the whole collection:
The main workflow skills use the practice skills in the Background group, such as code-cleanup, code-review, and pr. If you install only some skills, include those too.
2. In the repository you work on, type /setup-snappedly-skills in your coding agent. Setup asks about your code host, work tracker, and team workflow, and saves your answers in the repository.
3. Describe the change you want, then run the main workflow:
A person then reviews and merges the PR. To skip that review and let the agent merge the PR itself, end with /deploy instead of /submit.
To run a skill, type its name as a command: /grill. Some skills, such as setup-snappedly-skills, execute, clean-up, submit, and deploy, start only when you type their command. Others, such as grill, code-review, and tdd, load automatically when a task calls for them. If you are not sure which skill fits, type /help-snappedly.
The main workflow
Most work goes through four steps.
| Skill | What it does |
| --- | --- |
| grill | Interviews you about a plan until every open decision is settled. When the plan is too big for one PR, it proposes a spec with tracer-bullet tickets for execute to carry out as their blockers close. Skip it when the change is clear: describe what you want, let the agent propose a change, and run execute on that proposal. |
| execute | Carries out the settled plan, or a tracker issue you name, such as /execute #42. Name a planning spec to carry out its tickets whose blockers have closed. Substantial independent pieces, such as separate tickets, run in parallel agents. Then it cleans up, reviews the result, and opens a preview. |
| clean-up | The final check. For an independent look, run it in a fresh session, ideally on a different model or provider. It repeats cleanup and review over everything changed on the branch, fixes clear findings, and shows what it changed. Edits stay uncommitted. |
| submit | Commits pending changes, pushes, opens or updates the PR, marks it ready for review, and waits for its checks. A person then reviews and merges it: submit never merges. After review feedback, run it again to push fixes and re-request review. |
| deploy | Does what submit does, then merges the PR itself when your workflow allows, closes tickets, and verifies production if your workflow asks. It asks you first if commits from elsewhere or resolved conflicts entered the PR, and stops when the merge needs a person. PRs it plans to merge without review get the agent-merge label, so reviewers can filter them out. |
Use build-local when you want a browser preview of the current changes.
All skills
Each skill lives at skills/ / /SKILL.md, and the groups below follow the directories under skills/.
Begin: choose and configure the workflow
| Skill | What it does |
| --- | --- |
| help-snappedly | Picks the skill or workflow that fits your current situation. |
| setup-snappedly-skills | Records your repository's code host, work tracker, team workflow, domain docs, and frontend conventions for the other skills. |