If tigent used a wrong label when triaging, let me tell what it did wrong and have tigent improve its prompt via PR

[gr2m]

Example: vercel/ai#12223

It applied the ai/telemetry label. The reasoning for it is currently hidden in a GitHub Action build: https://github.com/vercel/ai/actions/runs/21639864959/job/62376020445#step:7:21

As a repository maintainer, I want to comment sth like this

@tigent The ai/telementry label was assigned incorrectly. This is an internal maintenance task. The ai/telemetry label should only be assigned if an issue is related to source code of the AI SDK.

As a result, @tigent should respond with its reasoning why it did what it did, and prepare a pull request to .github/tigent.yml or whatever file is used that documents how labels are assigned in the current repository. Right now it's hardcoded into the code:

app/app/api/webhook/route.ts

Lines 269 to 289 in aaea13e

	const { object } = await generateObject({
	model: config.model,
	schema,
	system: `you are a github pull request classifier. assign labels based on the pr content.
	available labels:
	${labellist}
	rules:
	- only use labels from the list above
	- pick labels that match the pr content and changed files
	- use label descriptions to understand what each label means
	- be conservative, only add labels you are confident about`,
	prompt: `title: ${pr.data.title}
	body:
	${pr.data.body || 'no description'}
	changed files:
	${filelist}`,
	});

Alternatively, I could ask why it did a thing first

@tigent The ai/telementry label was assigned incorrectly. Why did you do that

Which should result in @tigent to respond with its reasoning for that particular action. And then in a follow up comment ask it to update its prompt for labelling.

⚠️ we have to be careful about prompt injections if we let users collaborate on prompts.

[dancer]

resolved