Skip to content

docs(skill): search by meaning with the question, then with an example reply - #911

Merged
leemour merged 2 commits into
mainfrom
docs/skill-example-reply
Nov 18, 2026
Merged

leemour merged 2 commits into
mainfrom
docs/skill-example-reply

Conversation

@leemour

@leemour leemour commented Nov 18, 2026

Copy link
Copy Markdown
Member

Task

An agent that searches the owner's Telegram history passes the question to search conversations. A question reads
unlike the message that answers it, so answers in big groups ranked below the top 20.

Why

Measured on a test store (about 220,000 conversation chunks, OpenAI text-embedding-3-small), 10 questions with a
known answer, --limit 20. The example replies were written by a model that saw only the question.

Search with Answer in the top 5 Answer in the top 20
the question as asked 5 8
a one-sentence example reply 7 9
both, answer in either 8 10

The example reply lifted two answers by other people in big groups from outside the top 20 to 2nd and from 12th to
1st, and dropped one from 1st to 8th, so the skill keeps both searches.

What

  • SKILL.md step 1 and references/search.md: after the question, search by meaning again with a one-sentence
    example of the reply that would answer it, and read both lists.
  • Changelog fragment changes/docs-skill-example-reply.md.

Testing

  • pnpm docs:check.
  • Not measured: an agent run with the new skill.

🤖 Generated with Claude Code

@leemour
leemour merged commit e5021b0 into main Nov 18, 2026
14 checks passed
@leemour
leemour deleted the docs/skill-example-reply branch November 18, 2026 23:58
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant