Real uses of skill-engine, told honestly. Each entry is a record of one engagement where the engine did something specific — what the problem was, what the engine produced, what shipped, and what the author can and cannot prove. Every entry is sourced from real work, not composed from a representative-user composite.
-
Real uses, told honestly. Every case study is a thing that actually happened. No composite customers, no representative-user fictions. If the engine has not been used a way, that way is not in this series.
-
Numbers are measured, not estimated — or none. A figure appears here only if it was instrumented or recorded at the time. Defensible shape ("comparable modernizations ran multiple quarters; this shipped the following sprint") is permitted; reconstructed dollar math is not.
-
A tool other than this repo says so plainly. The first entry in this series describes work done with a domain-specific predecessor skill-engine I built inside one company — purpose-built for its codebase, not portable. This public repo is a separate, general-purpose tool built afterward from the same ideas. Cases that ran on the predecessor label it explicitly where the narrative introduces the tool; cases that ran on this public engine make no such caveat.
-
No composite customers. A case study describes one engagement, not a pattern abstracted across several. Pattern-level orientation ("if you're a forward-deployed engineer dropping into an unfamiliar codebase…") lives in
../usage-modes.md, not here. -
Recollection is permitted, and labeled. Some moments in these engagements were not instrumented — a teammate's offhand remark in a code review, the shape of a half-hour conversation. Where a beat is recollection rather than record, the prose says so. There is no shareable artifact behind these stories; the series rests on mechanism specificity and this contract, not on a showable receipt.
If you've used skill-engine for something the existing entries don't cover, open a case-study issue using the four-prompt template. Promotion from issue to published entry happens via a co-authored pull request — the truth-contract above is what you'd be agreeing to keep, and your name lands in the commit history alongside the maintainer's.