Discovery Needs Trust, Not Just a Registry

A registry answers "what tools exist?" ASPL answers a harder and more useful question: "which of these should I actually run, right now, given who vouches for them?"

The problem with a plain registry

Tool registries are lists. You search, you get matches, you pick one. But every entry looks equally legitimate — the malicious one and the battle-tested one render identically. For a human skimming a page, that's a minor risk. For an agent auto-selecting a capability in a loop, it's an attack surface.

Trust as a first-class ranking signal

In ASPL, every capability carries a trust score, and discovery ranks on a blend of intent match and trust:

combined = 0.7 * intent_score + 0.3 * trust_score

You can also set a floor. Ask for "must be at least 0.6 trust" and low-reputation matches are filtered out before you ever see them:

POST /v1/need
{ "intent": "summarize a legal contract",
  "min_trust": 0.6, "max_results": 5 }

Where trust comes from

Trust isn't a star rating someone types in. It's computed from the capability's confirmed track record: how many distinct agents acquired it and reported success, weighted by the publisher's own standing. A brand-new capability starts near zero and earns its way up.

Freshness matters too

Trust decays. A capability that hasn't been exercised in a long time shouldn't keep coasting on old reputation, so the score is decayed by staleness at read time. Active, working capabilities stay near the top; abandoned ones drift down on their own.

The difference is subtle but it changes the security posture: discovery stops being "here's everything" and becomes "here's what has earned a place in your shortlist."

None of these mechanisms are exotic on their own — reputation systems are well understood. What's useful is that the trust signal is wired directly into discovery and gated by the sybil-resistance described in a later post, so the ranking is hard to game.