The skill file is a photograph of your judgement

What a skill file gets right

Be fair to it: a standing-context file removes a genuine tax. Your conventions, your stack, the voice you want, the way you like things laid out — written once, read every session, never re-typed. For things that don’t change, a photograph is perfect: a portrait of a building that hasn’t moved is as true today as the day it was taken. That half of your working context is exactly what a file is for.

Where the photograph and the judgement part

The moment your thinking moves, the file is a picture of a past you. It records the pose, not the deliberation; the what, not the living why. There’s no shutter you press mid-session to say “keep this new call,” and the old frame stays in sharp focus beside the new reality until you hand-edit it. Three absences do the damage: no ratification, no supersession, no reasoning lineage.

No ratification

Nothing in a static file was put there by a gesture at the moment it mattered. You wrote it once, in advance, guessing what future-you would need. A decision you actually settle mid-work has nowhere to land except a later edit you may never make — so the file holds your predictions about your judgement, not the judgement itself.

No supersession

When you change a call, the photograph doesn’t retire the old one; it sits there, authoritative-looking, until you remember to overwrite the line. A stale entry in a skill file is read with the same confidence as a fresh one, because a snapshot has no way to mark which parts of it you’ve since moved past.

No reasoning lineage

A photograph shows the result, not the thinking. The file records “we do it this way”; it doesn’t carry why, what you ruled out, or what would change your mind. Hand a model the conclusion without the reasoning and it fills the gap with a rationale of its own — and a bare rule invites it to reopen the very question you closed.

What the living thing looks like

Unl holds the judgement, not a photo of it: decisions entered on your word, superseded when your thinking moves, each carrying its reasoning — served as unprompted relevant context the moment one bears on the work. You stay in command of what’s held: saved on your word, retired on your word. Tuned for Claude, Claude Code, ChatGPT & Cursor at launch, extending across the AI ecosystem. Connects anywhere MCP does.

This is one half of a pincer. The static file freezes your judgement at a snapshot; the other half is a loop that acts with no settled decision in reach at all. The loop-counterfactual pages describe that side; together they are the two ways your context fails a model — frozen, or absent.

A skill file is a true photograph of your judgement the day you write it; what a photograph can’t be is the judgement itself — kept current, superseded when you change your mind, carrying its why — so the answer comes back measured against what you already decided, and why.

Reads through Unl arrive with measured context — in the presence of the decisions you’ve already settled. The reach lane is live: one box, paste anything. If it speaks MCP, Unl can reach it. Readings arrive unprompted, the data beside the criterion; Unl is a courier, not a warehouse, and keeps only your keys and the frame.

Read further

• Why static context goes stale
• Hand-maintained context vs a substrate
• Static context solves repetition, not drift
• What the agentic loop diagrams leave out
• Unl vs the rules-file method