fix(evi): drop the linear mcp connection and raise the session limits - #512
Conversation
|
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
|
Thank you for following the naming conventions! 🙏 |
commit: |
|
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: Organization UI Review profile: ASSERTIVE Plan: Pro Plus Run ID: 📒 Files selected for processing (7)
📝 WalkthroughWalkthroughThe changes increase agent session token limits, remove the Linear MCP connection, document OAuth failure behavior, revise runtime metadata documentation, and add guidance for concise comments. ChangesAgent configuration and documentation
Estimated code review effort: 2 (Simple) | ~10 minutes Possibly related PRs
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Two changes, both from what the EVL-213 run showed.
The Linear MCP connection is removed
It does not degrade when misconfigured — it takes the run down. On that thread every GitHub tool returned
Cannot read properties of undefined (reading 'toLowerCase'), thrown fromisProvisionableConnectorUidin@vercel/connect/dist/eve/provision-oauth-connector.js, the OAuth connector provisioning path this connection uses. Five calls, same error, and Evi finished the task by reading the repo over HTTP from the sandbox instead.Local evals never caught it:
provisionEveOAuthConnectorreturns early without an OIDC token, so the code only runs deployed. I said earlier that a broken connection would just mean "Evi answers without Linear" — that was wrong, and this is the correction.The Linear channel stays. Only the MCP connection goes, until Connect can mint an app token for that connector.
Session limits raised
5Minput was set defensively before there was anything to measure. That thread spent 2.8M input across two turns, so a conversation was about three turns from being cut off mid-thread. Now20Minput /250koutput — still half of eve's default, and it bounds a loop rather than a long conversation. Cost is not what these protect: that session billed $0.0103 at a 95% cache rate.Also
AGENTS.mdgains one line under "Code style — no slop": a comment states a durable constraint in one or two lines, and the investigation that produced it belongs in the PR body or a doc. Applied to the threeagent/libfiles here.Testing
pnpm eval16/16, 86 gates.eve inforeports 0 errors, connections down todocs, channels unchanged. Lint and typecheck clean.Summary by CodeRabbit