Observed skills vs claimed skills

Most skill profiles are built from what people say they can do. A developer writes "React, TypeScript, system design" on a resume, and that list stays frozen until someone manually updates it.

The problem is not dishonesty. It is staleness and inflation. People add skills after a single project and never remove them. They list technologies they touched once. Over time, the gap between the profile and the person widens — and nobody notices, because there is no feedback loop.

Static skill tags are also impossible to compare. When two people both claim "senior-level system design," there is no way to know what that means in practice. One might architect distributed systems daily. The other might have drawn one diagram last year.

Temet takes a different approach. Instead of asking what you can do, it watches what you actually do.

When you work with an AI coding agent — Claude Code, Codex, or any tool that leaves session traces — Temet reads those sessions and extracts the skills you demonstrate. Not the technologies you mention, but the behaviors you repeat: how you scope work, how you prompt, how you manage context across long threads, how you make trade-off decisions.

Each skill gets mapped to a proficiency level (from novice to expert) based on frequency, consistency, and quality of evidence. The result is not a tag. It is a structured competency with examples, anti-patterns, and progression data attached.

This is what machine-attested means: the observation comes from real interactions, not from a form.

A single audit produces a snapshot. But Temet is designed for the long game.

Every time you run a tracked audit, Temet compares the new result against your previous profile. Skills that appear consistently get stronger evidence. Skills that fade get flagged. New patterns emerge as your work evolves.

When you publish a public profile, this progression becomes visible. Other agents and peers can read your profile and see not just what you know, but how your skills have changed — and whether the trajectory is going up.

This is fundamentally different from a LinkedIn endorsement or a static badge. The evidence is machine-readable, versioned, and tied to real work. It compounds because each session adds signal, and the profile updates to reflect reality.

Trust in professional skills has always been social: recommendations, interviews, referrals. These work, but they do not scale and they are hard to verify.

Machine-attested skills introduce a new layer. When an agent reads your Temet profile, it does not need to trust your word. It reads structured evidence: specific examples, observed patterns, proficiency levels backed by session data. It can calibrate its behavior — acknowledging your strengths, watching for your known blind spots, suggesting next steps that match your actual level.

This is not a replacement for human judgment. It is a foundation that makes human judgment more informed. The profile is always reviewable by the user, always under their control, and always grounded in work that actually happened.