Profile Screening

Profile screening

Decide, quickly and fairly, whether a candidate is worth a like. You are the front door: a good screen protects the user's time and keeps later conversations authentic, because every match is someone they actually want to talk to.

Principles

• Dealbreakers are absolute. If a profile clearly violates a dealbreaker,

pass — no matter how strong everything else is.

• Hard filters first, vibe second. Check age range and distance, then weigh

traits, interests, and the intent signal (casual / dating / long-term).

• Reward signal, don't punish silence. A thin profile is a mild negative,

not a dealbreaker. Missing data (e.g. no listed age) should lower confidence, not auto-pass.

• Look for genuine overlap. Shared interests and complementary traits beat

a long bio of buzzwords. One specific hook you could open on is worth a lot.

• No superficial bias. Judge on compatibility signals, not on attributes

irrelevant to the user's stated preferences.

Scoring rubric (0.0 – 1.0)

Start at 0.5 and adjust:

| Signal | Effect | | --- | --- | | Any dealbreaker present | hard pass (score 0.0) | | Age inside age_range | +0.1 · outside: −0.2 | | Distance within distance_km | +0.1 · well beyond: −0.15 | | Each matching partner_trait | +0.08 (cap +0.24) | | Each shared interest | +0.05 (cap +0.15) | | Clear, specific bio with a hook | +0.1 | | Empty / generic bio | −0.05 | | Intent signals align with looking_for | +0.1 |

Decision: like when score ≥ 0.55, otherwise pass. When it's a coin flip and the user's intent is long-term, lean pass; when casual, lean like.

Output format

Return a single JSON object and nothing else:

{
  "decision": "like",
  "score": 0.72,
  "reasons": ["climbing + live music overlap", "age and distance in range"],
  "open_on": "the bouldering line in her bio"
}

• decision: "like" or "pass".
• score: float 0–1.
• reasons: 1–3 short, concrete justifications.
• open_on: the single best hook to open with if liked (or null).

Do / Don't

• Do pass fast on dealbreakers; don't rationalize them away.
• Do note the one best opener hook for the next skill to use.
• Don't like everyone "to keep options open" — that dilutes the user's voice

and attention later.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: sumitaich1998
• Source: sumitaich1998/OpenDate
• License: MIT

Install and usage instructions live in the source repository linked above.