@jehudarajasa/agency-skills weekly downloads statistics

Skills for Everyday Software Engineering

Coding and execution can be delegated to agents. The agency is still yours.

Jehuda's daily driver Claude Code skills marketplace, shaped by more than five years of engineering experience at agencies.

Agency Skills is built for issue-driven development: pull an issue from any issue tracker (GitLab, GitHub, Jira, or Linear), cut a branch, plan, implement, review, commit, and open a change request.

Spec Kit and BMad are popular spec-driven development frameworks, but they introduce more process and context bloat than the typical engineering workflow needs. Not every issue warrants a spec, and engineers working from a groomed backlog do not need product discovery or UX design skills in the same toolchain. Agency Skills are human-invocable only, meaning they do not bloat your model's system prompt and agents cannot invoke them implicitly. Simple, minimal, and built with YAGNI philosophy in mind.

I am a fan of Matt Pocock's skills. In fact, I built Agency Skills using his /writing-for-agents skill. These skills were created to fill gaps I encountered in my day-to-day work as a product engineer. See how Agency Skills overlap with Matt's Skills.

Installation

Agency Skills depend on Matt Pocock Skills for /grilling, /to-tickets, /tdd, and /code-review.

Claude Code and Codex install both Agency Skills and Matt Pocock Skills as plugins. OpenCode installs Agency Skills as a plugin and Matt's Skills through the skills CLI. Other harnesses install both skill sets through the skills CLI.

<details> <summary><big><strong>Claude Code</strong></big></summary>

Run from your terminal:

npx @jehudarajasa/agency-skills@latest install claude

</details> <details> <summary><big><strong>Codex</strong></big></summary>

npx @jehudarajasa/agency-skills@latest install codex

</details> <details> <summary><big><strong>OpenCode and Other Harnesses</strong></big></summary>

npx @jehudarajasa/agency-skills@latest install opencode

Replace opencode with your agent's name.

</details>

Setup

Run /setup-agency-skills (or $setup-agency-skills for Codex) once per project to detect your issue tracker and map your board's workflow states into .agents/config.yml (see config.example.yml).

Workflow

1. /fetch-issues

List your open issues in the project's issue tracker so you can pick one.

2. /cut-branch

Cut a branch named after the issue in <type>/<issue-id>-<slug> format.

3. /plan-implementation

This is the load-bearing skill within the workflow, where you will likely spend most of your time. The generated plan is the agreed contract between you and your agent.

Issues often come off the board thin—a title, a sentence, whatever grooming left behind. This is where you and the agent turn that into a concrete plan, and /grilling grills it out first when a real decision is still open. The plan is yours; the typing is the agent's. Plans follow the issue workspace layout under .issues/<issue-id>-<slug>/. See why plans are persisted locally.

When the issue's intended behavior, affected scope, and verification are already clear, go straight to /implement-change. It follows a saved plan when one exists; otherwise it works directly from the issue. Planning is driven by unresolved decisions, not the number of files involved.

4. /implement-change

Implement from a saved plan or clear issue requirements. In repositories that already have tests, implementation runs through TDD. This skill tells your agent to verify and stop before review and commit. See why implementation stops before commit.

5. /code-review

In this step, you can review the agent's work by hand, with Matt Pocock's /code-review, or both. The skill reviews the changes against two independent axes: the repository's documented standards and the originating issue or plan.

Skip directly to /commit for trivial changes that do not warrant a full code review, such as comment edits or typo fixes.

6. /commit

Create small Conventional Commits, one concern each.…