Waste
The waste catalog is in the published standard (muda standard). Each recipe is a known fix for one or more classes. The ci pack's fourteen recipes cover GitHub Actions, CDK, Docker, Python, Node and Go, and the aws pack covers waste in AWS accounts (muda recipes).
waste classes
| Class | Waste | What it looks like | Recipes |
|---|---|---|---|
| W1 | No-op not a no-op | An unchanged deploy still updates stacks or builds images. | R2 R3 R4 |
| W2 | Unpinned toolchain | Floating versions let the same tree produce different outputs. | R1 |
| W3 | Emulated builds | Images build under CPU emulation instead of natively. | R6 |
| W4 | Uncached work | Unchanged installs or layers are rebuilt, including a cache that never warms. | R6 R11 |
| W5 | Serial where it could be parallel | Independent work runs one at a time, or the same computation runs twice. | R5 |
| W6 | Re-verification | The same checks run more than once on the same tested tree. | R10 |
| W7 | Chain stalls | Automated hand-offs fail or stall with no visible outcome. | R7 |
| W8 | Silent runner-plane failure | Jobs queue because the runners cannot take work, and nothing alerts. | R8 |
| W9 | Rebuilt artifact | Identical inputs are built again in a later stage. | R6 |
| W10 | Ignored advance notice | CI warnings about platform or toolchain changes go unread. | R9 |
recipes
Each recipe gives the symptom, how to detect it in real logs, the fix, its traps and how to verify it, with detail for each stack it covers. Recipes come in packs you buy on subaud.tools; muda syncs the packs your account owns with muda packs sync. The ci pack holds these:
| Recipe | Fix | Waste |
|---|---|---|
| R1 | Pin the toolchain through version files | W2 |
| R2 | Turn off framework version reporting | W1 |
| R3 | Keep rotating secrets out of image asset hashes | W1 |
| R4 | Isolate images that share a build context | W1 |
| R5 | Find the duplicate before parallelizing tests | W5 |
| R6 | Build once, outside deploy; promote by digest | W3 W4 W9 |
| R7 | Automated hand-offs never stall silently | W7 |
| R8 | Keep runners current and make a refused plane loud | W8 |
| R9 | Read and act on advance notices | W10 |
| R10 | Reuse proven checks without dropping deploy-specific gates | W6 |
| R11 | A cache that never warms | W4 |
| R12 | Split a service rollout into phases before tuning it | W5 |
| R13 | Key the dependency layer on third-party dependencies only | W4 |
| R14 | Deploy contracts first, then every stack at once | W5 |
ask your agent
The standard ships inside muda and the recipes come from the packs you have synced, so your agent reads exactly what you run. Ask in plain words; each example shows the kind of answer you get back.
› What does muda's waste class W4 mean, and how would we detect it here?
W4 is uncached work: unchanged dependency installs or layers are recomputed. To detect it, muda repeats identical inputs and checks the real hit or miss evidence for each asset and layer. A common cause is a rotating build argument, which invalidates every layer after it.
› Show me muda's recipe for a cache that never warms, for a Go repo.
R11, a cache that never warms. Symptom: a setup or build step stays slow on every run although its dependency cache reports a hit. For Go, setup-go's cache is on by default, so a light job with no cache input can still save the shared key first. The fix: only the job that builds the full dependency set saves it; turn the cache off in the others.
› Which muda recipes apply to this repository's stacks?
This repository is Go on GitHub Actions. Four recipes have Go detail: R1 pin the toolchain through version files, R5 find the duplicate before parallelizing tests, R6 build once and promote by digest, and R11 a cache that never warms.