Hacktoberfest 2026: the issues maintainers tagged for October, open and beginner-friendly. Browse Hacktoberfest issues

The standing fit check leans on the model's confidence alone

Open
  • Is something wrong?
#490 0 comments 0 reactions 0 assignees View on GitHub

Maintainers usually reply within 1 day

@Sanketmis208 is already working on this.

Since Oct 20, 2026.

  • #578 by @Sanketmis208 — open

Assessment

Difficulty
5/5
Estimated time
Over a week
Newbie friendliness
22/100
Issue type
Feature
Clarity
Mostly clear
Activity status
Active
Tech stack
typescript
Domain
ai-infra-agents

Research direction

Start from the quick model check that reviews each email sent under a standing approval, in the hive area (memory, routing, GOD agent); the issue does not name the file. Read how the approval and email are passed into that prompt. The proposal adds deterministic checks for numbers, dates, names and negation, plus a per-item model question, so it is design work. Done means hard-check failures always go to the owner and the prompt stays byte-identical when no item list is supplied.

Written by the indexing model from the issue text.

Description

enhancement
Problem / motivation

A standing approval lets an agent send one kind of email without asking the owner
first. Every send under it is checked by a quick model that reads the approval and
the email and answers whether it fits. That check is the only thing between an
email that goes further than the owner allowed and the owner's inbox, and it rests
entirely on the model's own confidence. If the model overlooks a fact, invents a
date, adds a price, or drops a "not", the email goes out without the owner seeing
it. There is nothing countable to check against.

Proposed solution

Put a countable check behind the model, and keep the model for the part that
genuinely needs judgement.

  • Pull the closed classes out of the request: amounts and numbers, dates (with
    relative ones like "Friday" or "next week" flagged as unresolved), names, and
    negation. Check these without the model. If the email introduces a number,
    date or name that is not in the request, or leaves a relative date unresolved,
    it fails closed and goes to the owner.
  • Pull the open classes too, with the same pure-JS part-of-speech library: nouns
    with their leading adjectives, adverbs with their nearest verb. Give the model
    those items as a closed list and ask it one question per item (present,
    paraphrase, or absent) plus the kind fit, instead of an open "does this fit".
    The model may rescue an item it recognises as the same thing, but it can never
    pass an email that failed a hard check.
  • Keep the current prompt byte-for-byte when no list is supplied, so nothing
    changes until this is switched on.
Screenshot, mockup or recording

No response

Alternatives considered
  • Replacing the model's judgement with pure word matching. It would miss
    paraphrases and cannot verify a negation, so it would send the wrong things in
    the other direction. So, replacement doesn't make as much sense as collaborative
    scoring.
Area

The hive (memory, routing, GOD agent)

Pre-flight
  • I attached a screenshot, mockup or recording above.
  • I searched existing issues and discussions, and this isn't a duplicate.
Dominant language
TypeScript
Stars
273
Forks
211
Avg merge
1d 22h
Merged PRs (30d)
316

Getting set up

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

More from agentvivekkumar/dontbemichael

All issues in agentvivekkumar/dontbemichael

Similar issues

More TypeScript issues

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.