Just-in-time expertise
SEIF makes the AI a just-in-time expert in its own capabilities, so the human never has to be. Surfacing a compact capability map lifted a cold model's verb-reach from 0.50 to 1.00 — measured.
The point of SEIF is not that you learn forty CLI verbs. It is that the AI working beside you reaches the right one at the moment of need — so you never have to be the expert in SEIF to get its value.
That is a claim about discoverability, and we measured it.
The gap: capability without salience
The capabilities already ship and already work. The failure mode is subtler: a fresh AI session, faced with a real need, does not reach for the verb that solves it. It confabulates a plausible-but-wrong command, or falls back to "I can't," when the exact tool was one line away.
The example that started this: an agent insisted it could not see a peer agent's
progress — when seif relay watch and seif relay poll exist precisely for that.
Not a missing capability. A discoverability and salience gap.
The fix: an intent-indexed capability map
SEIF surfaces a compact, intent-indexed map at session start — need → verb, grounded in the live CLI surface. Not documentation to read; a reach-for-the-right-thing index the model already has in context when the need arises.
Track / watch a peer agent's progress →
seif relay board·relay watch·relay pollRemember a durable fact across sessions →seif gov memory addRecall a past decision ("where did we decide X") →seif memory queryFind relevant code without reading whole files →seif context query·context retrieveClassify content sensitivity before sharing →seif classify
That is a sample. The complete, scannable corpus — every need → verb pair plus the multi-step workflow recipes — lives on the Intent → verb map.
What we measured
We ran a cold-AI A/B test on a frontier model (Claude Sonnet): ten real needs, phrased differently from the map (testing generalization, not echo), each run in a clean environment with no SEIF hooks — with and without the map in context.
| condition | verb-reach |
|---|---|
| No capability surface | 0.50 (5/10) |
| With the capability map | 1.00 (10/10) |
A +0.50 lift. Without the map, the model confabulated wrong-but-plausible verbs; with it, it reached the exact correct verb on all ten. The lift replicates against the produced session-start surface, not just a hand-written sheet — so it is the real shipped form that carries the gain.
Mechanism, not numerology
This is the inverse of the resonance-math result: there, front-loaded framing hurt. Here, a small surface of the right information at the right moment helps — and it helps for an ordinary, inspectable reason. The signal is clear; with n=10 the exact figure is noisy, and it proves need → verb reliability, not discovery of unknown verbs.
Why it matters to you
If you have ever re-explained the same tool to an AI every session, that is the product gap — and it now has a measured fix. SEIF's job is to make the assistant fluent in SEIF on your behalf: the human states the intent in plain language, and the AI reaches the verb that does it. Just-in-time expertise, so you don't have to carry it.