Output becomes input becomes improvement
Why chains beat isolated skills
A single skill is a useful artefact. A chain of two or three skills, each feeding the next, is a substantively different category of object. The first skill produces something — a draft, a log, a brief — and the second skill operates on that something, and the third operates on the second's output. The chain compresses what would otherwise be a session of manual stitching into one command, and it captures the working sequence in a form that can be reasoned about as a system. The transition from one skill to a chain is one of the cleanest jumps in personal productivity.
The three-skill template
The chains that working setups converge on tend to have the same shape. A first skill captures or generates raw material — a log of what happened today, a draft of a deliverable, a set of variants. A second skill analyses or distils that material — pulling patterns from the log, editing the draft against an objective, selecting from the variants. A third skill improves the system itself — updating the knowledge base, adjusting an earlier skill against what was just learned, surfacing a pattern that should drive a future change. The shape is recognisable across very different content areas, which is the usual sign that the structure is real rather than incidental.
What makes a chain compound
A chain that runs once produces a result. A chain that improves its components on each run produces compounding. The cleanest mechanism is to have the third skill in the template write its findings back into the system the first two skills read from: the knowledge base, the project instructions, the skill files themselves. Over weeks the patterns the chain has noticed are baked into the upstream. The first skill becomes better at capturing what matters because the system has learned what mattered last time; the second skill becomes better at distilling because it knows the kinds of distillation that paid off. Compounding here is not a metaphor; it is the visible result of writing improvements back into the upstream artefacts.
Honest limits of chains
Two warnings are worth making explicit. The first is that chains amplify both improvements and errors. A bad early skill in a chain is a bad input to every later one, and the errors compound just as cleanly as the corrections. The discipline of getting the early skills right before extending the chain is what stops this. The second warning is that long chains become brittle. A chain of six skills is hard to debug and harder to reason about; the working compromise is to keep chains short, three skills if possible, four at the outside, and to keep a clear human checkpoint at one of the steps. The chain is a tool, not a destination.
Of the recurring routines you perform, can you name one where the output of step one feeds step two, the output of step two feeds step three, and the result genuinely benefits from each handoff?