Agent skills
Skills for AI coding agents, each with what it costs to load, how much of it is scripts and what it touches on your machine.
How to read a skill
What it costs to keep
Idle in context, in use when it runs, and up to with its linked files.
Scripts or instructions
Script-led, mixed or model-led: how much the model has to work out itself.
What it can reach
Shell, network, file writes, git and credentials, declared or detected.
Quality, security, repo health
Out of 100, each with what it is made of. Turn a card over to see them.
Guardrail Hooks
687Agent Workflows
Turn a rule an agent keeps breaking into a permission rule, hook or script, with templates for a one-way-door guard and a commit-identity guard.
Handover
687Agent Workflows
Write the one note a fresh agent session starts from, after checking the disk, the running jobs and git against what the session remembers.
independent-review
687Code Review And Quality
A release gate built on a fresh, read-only reviewer that never saw the build: git-computed subjects, a five-part brief, mandatory triage and a clear stopping rule.
intended-vs-implemented
687Code Review And Quality
A review method that compares AI-built code with its recorded intent to find unenforced permissions, public cron endpoints and private data leaking onto public surfaces.
Retro
687Agent Workflows
At the end of a session, list the friction, find its cause and build a checked fix into your agent setup, or conclude that no change is warranted.
start-here
687Agent Workflows
Say what you have (an idea, a change, a bug or a release) and get the product's next step and the skill that does it.
test-harness
687Code Review And Quality
Rules for tests where a green gate means something: what each harness fakes, how to run and report the gate, regression tests for fixes, and lessons from real bugs.