Measure

Risk class

You might say: risk tier, blast radius, door type

A label for how much damage a change can do and how hard it is to undo. A change that only composes existing pieces is low risk. One that extends a piece's behavior or contract is medium. One that adds a piece, a data store, an outside service or a trust boundary, or that crosses a one-way door, is high. The class decides which gates the change must pass and whether an agent may merge it.

What it does

Sends low-risk work through trusted automation and keeps human attention for high-risk and one-way-door changes.

Ask

Does this change compose, extend or add, and how costly is it to undo?

What to look for

1. A low, medium or high label on a pull request
2. A pull request summary that says which pieces the change composes, extends or adds
3. High-risk pull requests waiting for a named human

Catalogue: Risk tier

• Low: The change composes existing pieces as designed. Green checks are enough to merge it. The change uses pieces that already exist and does not change how they behave.
• Medium: The change extends an existing piece's behavior or contract. It needs green checks, an AI reviewer and the run's own proof that the change works. The change alters how an existing piece behaves, so its callers can break.
• High: The change adds a new piece, data store, outside service or trust boundary, or it is a one-way door such as a public API name. A human approves it before merge. The change is costly to undo or widens what the system can reach.

Pattern

Two-way door

Connections

• Controls: Merge policy

Sources

1. Kody, kody-hq/kody repository, docs/contributing/architecture/primitives.yamlwires existing primitives without changing them … changes a primitive's behavior, shape, or contract … introduces a new primitive
2. Kent C. Dodds, kcd-skills, autopilot skillP2 Risk tiers: low = CI green; medium = +AI reviewer +rung 4 proof; high = +human
3. Kent C. Dodds, You Own the Outcome: Product Engineering in the Age of AI, Design the playgroundAutomate shipping where the risk is low or medium. Well-structured PRs, tests, and AI reviewers let a lot of changes ship without you in the middle. Save your attention for the high-risk, one-way-door stuff.
4. Addy Osmani, Agentic Code Review, Should we just let AI review more of it?How much human you keep is a dial, and you set it by blast radius, not by guilt.
5. Birgitta Böckeler, To vibe or not to vibe, A sliding scale of review effortHigh probability + high impact + low detectability High level of review is advisable. Assume the AI might be wrong and cover for it.

Use it on your project

teardown-apply: hold "Risk class" (risk-class) from the teardown at https://teardownatlas.com/software-factory/ against my project

Paste it into a coding agent that has the teardown skills.

An AI agent wrote this map with the teardown skill. It can contain errors: check the sources before you rely on a part. About and legal