Match client requests to the right open issue
Each week, have Jev match every client request to an existing GitHub issue, sending low-confidence matches to a person.
Author's figure: Author's figure: matched Opus accuracy on 40 requests, decisions 150x cheaper, 5 bad matches stopped by reviewers (Akshay Pachaar). Reported by the builder, not measured by us.
The use case
Agencies and product teams collect client requests from email and calls, then lose time working out whether each is already tracked. This agent takes each request and asks Jev to choose the matching GitHub issue from the open list, or none. Confident matches get linked and the client gets a status update drafted from the issue; low-confidence ones go to a person. In the author's test, reviewers still caught a handful of wrong matches, which is why the review step stays. A project manager approves any message to the client.
Example workflow
- Collect client requests from email or a form into a weekly batch
- Fetch open issues from GitHub
- Ask Jev a Choice question over issues plus a none option
- Link confident matches and queue low-confidence ones for review
- A project manager approves client updates before sending
Tools
- Jev (TypeSafe AI)
- GitHub
- TrueFoundry
- Daytona
Risks and guardrails
- Similar issues cause wrong links
- Long issue lists may need pre-filtering
- Client updates sent from a wrong match damage trust
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Jev inside an open agent harness by Akshay Pachaar on Made with Jev: madewithjev.com/builds/jev-in-trueforge-agent-harness
- Jev inside an open agent harness: original post or repo by Akshay Pachaar: x.com/akshay_pachaar/status/2109909532352917658
- GitHub MCP server (official page): github.com/github/github-mcp-server (checked 2026-11-09)