skill-quality-suitelisted

# Skill quality suite Validates a `SKILL.md` against the Agent Skills specification, lints the instructions an agent will actually follow, scans a skill for secrets and prompt injection before you install it, checks whether it will work on another agent, and measures whether it improves the agent's work at all. A quality gate in two halves. The **static** half reads the skill: eight checks, one command each, separate because they fail at different moments - structure breaks today and in silence, the specification breaks on publication, compatibility breaks on somebody else's machine, routing breaks when a neighbour's description moves. Running them as one undifferentiated pass reports all four with the same urgency, which is how a report stops being read. The **evaluation** half runs an agent: does the skill fire, does it make the work come out better, did the last edit make it worse. It costs money and minutes and needs an agent installed, so nothing in it runs unless you ask for it by name. Everything lives in `scripts/sqs.py`. The static half is standard library only and works offline; it finds the skills folder on its own, or takes `--skills-dir`. ## Paid layers: offer, warn, never start `sqs.py eval` in every form, and any check a model has to read or judge, spends the user's money or the usage window of their subscription. The rule for all of it: 1. **Never run a paid layer on your own initiative**, however well it fits the task. Run it only after the user sai