Aidan's Agent Skills

The skills I use for my agentic development workflow.

Spec out the requirements with a grilling session, create GitHub Issues and PRDs for auditability, run your agents autonomously, keep yourself in the loop.

Quickstart

Run the installer and add the skills to your coding agents:

npx skills@latest add chieaid24/skills

Then, in a repo you want to work in:

• /bootstrap-issues sets up the queue.
• /spec shapes an idea into a PRD with child issues, or straight into one or a few issues for a small change.
• /start-next-issue works the queue.
• /catch-up reviews what happened.

To work several repos in one run, invoke /start-next-issue from the folder that holds them (say ~/projects) with a .agent-repos file listing the clones, one per line. It then picks each issue from whichever repo has the most-blocking one.

Skills

All eleven are model-invoked: the agent can reach for one on its own when the task fits, or you can type it.

The issue-queue workflow

• bootstrap-issues - One-shot repo setup for the queue: CI test gate, work-queue labels, self-merge branch protection, issue template, pre-commit hooks, and a grilled DESIGN.md for frontend repos.
• setup-pre-commit - Stack-aware pre-commit hook that runs format, lint, and test before every commit.
• spec - Grills a rough idea into sharpened terms and ADRs, then publishes either a [PRD] with child slices or a few dependency-linked issues.
• start-next-issue - Claims the most-blocking ready issue, drives it to a merged PR, and chains up to three of them, halting on any failure. Runs against one repo, or against every repo in a list when you invoke it from the folder above them. Never touches hitl.
• catch-up - Read-only morning report of what shipped, what stalled, and which hitl issues are waiting on you.

Autonomous audit passes

• code-audit - Deepens shallow modules through subagents, reverting any change the unit, integration, and E2E suites cannot prove behaviour-preserving.
• ui-audit - Screenshots every flow, measures visual defects and design-system drift, and fixes each with a committed probe that was red before and green after.
• security-audit - Multi-agent hunt for exploitable vulnerabilities: recon, attack-class hunting, adversarial validation, and a schema-checked report.

Driving the browser

• drive-browser - Drives your already-signed-in Chrome through the chrome-devtools MCP tools so an agent finishes login-gated web steps on its own: starts or reseeds the debug browser, never types passwords, and gates irreversible actions.

Writing and meta

• writing-great-skills - The vocabulary and principles for writing skills that behave predictably, backed by a full GLOSSARY.md.
• writing-style - House style for anything a reader outside the work sees, tuned to fight a model's prose defaults.

Credits

code-audit and writing-great-skills are adapted from mattpocock/skills, and security-audit from cloudflare/security-audit-skill, all MIT-licensed.