Claude, Launch Review.
Uses Claude's launch-review skill to show that walking through a launch checklist is a fixed specification Claude follows, not evidence Claude has authority to approve a launch.
Claude has a skill called launch-review, and the name raises an obvious question: does that mean Claude can approve a launch? No, and the gap between what the name suggests and what the skill actually does is the whole point worth understanding.
What a skill actually is
A skill is a folder holding one file, SKILL.md, that Claude reads before it starts and then follows step by step, in order. Delete the folder and Claude loses no launch authority, because there was none to begin with. It just stops following that one checklist. Launch-review is a folder holding a single SKILL.md with instructions, in plain language, for walking through a launch checklist step by step.
How the checklist runs
Claude reads that file top to bottom and works through it step by step in the order it's written, branching only where the file itself says to branch. That makes a skill a specification, not new judgment. The payoff is the same checklist applied to every launch, every time. The limit is that anything outside those written steps, and Claude has no special opinion about it.
What launch-review doesn't grant
Launch-review never gave Claude authority to approve anything. It guarantees only that every time the skill runs, Claude reads the same file and follows the same checklist. That's the whole mechanism. Reading the skill's name as sign-off authority is a natural misread, since "review" carries a connotation of judgment in common usage, but the actual behavior is closer to running a fixed inspection than rendering a verdict.
What the output does and doesn't prove
Watching Claude flag every risk in a launch checklist doesn't prove it understood the launch in any deep sense. A checklist can be followed to the letter and still miss something it was never asked to check. Watching Claude flag nothing doesn't prove the launch is clean either; it may just be a case the checklist doesn't cover. Neither outcome is proof one way or the other, because the skill's job is procedural, not evaluative.
Trying it yourself
Pick one decision you check the same way before every launch, before a feature ships, before a claim goes out. Ask Claude to write a SKILL.md for it, plain language, ordered steps, then have it read the file back to you and walk through exactly what it will check before checking it. Running this against a real recurring decision shows the mechanism directly.
Key takeaways
- Claude's launch-review skill is one plain-language SKILL.md describing a fixed launch checklist.
- Claude follows the file's steps in a fixed order, branching only where the file explicitly says to.
- A skill is a specification Claude executes, not authority to approve or greenlight anything.
- The same checklist runs for every launch, every time the skill executes.
- Flagging every risk, or flagging none, proves nothing on its own about whether the launch was actually understood.
Who this is for
Product and legal teams evaluating whether Claude can help standardize pre-launch review, and anyone who wants a precise picture of where a Skill's checklist ends and human sign-off authority begins.
Full transcript(auto-generated, with timestamps)
[0:00]You guess Claude approves a launch once it picks up a skill called launch review. It doesn't. It works through a written checklist. Let's see what's actually inside that file. Here Claude has a launch review skill, and it sounds like Claude itself got sign-off authority, like it can greenlight a launch and no one needs to check further. But nothing in the model changes. Delete the skills folder and Claude doesn't lose any launch authority. There was none to begin with. It just stops following that one checklist. Here's what a skill actually is, one file. Launch review, for instance, is a folder holding a single skill. MD instructions
[0:35]For walking through a launch checklist step by step, written in plain language Claude reads before it starts. Claude reads that file top to bottom and works through it step by step in the order it's written, no branching unless the file itself says branch. That makes a skill a specification, not new judgment. The payoff, the same checklist every launch every time. The limit, anything outside those written steps and Claude has no special opinion about it. So, launch review never gave Claude authority to approve anything. It just guarantees that every time launch review runs, Claude reads that same file and follows the same checklist. That's the whole trick. Watching Claude flag every
[1:13]Risk in a launch doesn't prove it understood the launch. A checklist can be followed to the letter and still miss what it was never asked to check. And watching it flag nothing doesn't prove the launch is clean. It may just be a case the checklist doesn't cover. A skill named launch review doesn't hand Claude authority to approve a launch. It's a checklist Claude reads before it starts, so the same steps run every time and anything outside those steps is still on you. Your turn. Paste this into Claude. Pick one decision you check the same way before every launch, before a feature ships, before a claim goes out. Write me
[1:44]A skill MD for it, plain language, ordered steps. Then read it back to me and walk me through exactly what you'll check before you check it. Claude launch review lay 'em in for bear.
