docs: require a user flow and a stuck-at proof in feature requests

[devin-ai-integration]

TLDR

Problem this solves:

• Feature requests arrive as abstract prose, not user tasks
• Nothing shows the requester actually hit the wall
• Undocumented features get filed as missing ones

How it solves it:

• Swap "Motivation, pitch" for a before/after this feature User Flow
• Add "How far you got": live-proxy attempt up to the dead-end
• Guidance lives in descriptions, so nothing ships prefilled
• Triage now requires evidence of the dead-end
• Agent Shin's close and grace comments ask for that same evidence
• Dead-end proof a requester did paste gets credited back

Stacked on #36498 (now merged), which did the same for bug reports

User Flow

Before: someone requesting a feature gets two free-text prose boxes, so maintainers get a wish rather than a task, with no evidence it is actually missing

1. They open https://github.com/BerriAI/litellm/issues/new?template=feature_request.yml
2. "The Feature" and "Motivation, pitch" both invite abstract prose, so they describe the knob they imagine instead of the request they want to send
3. Nothing asks them to try it, so a feature that already works under a different parameter name lands as a request, and a real gap lands with no reproduction
4. A maintainer reads the issue and cannot tell what call should start working, or where it dead-ends today

After: the same form asks for the flow with and without the feature, then for the run that dead-ends

1. They open https://github.com/BerriAI/litellm/issues/new?template=feature_request.yml
2. "User Flow" shows its rules above the box and a worked example as ghost text, and asks for "Before this feature (today)" and "After this feature (ideal user flow)" lists in the end user's words, with URLs, payloads, and what visibly came back, "Before this feature" also carrying the workaround they live with
3. "How far you got" asks them to walk the "After this feature (ideal user flow)" list against a live proxy they ran, hitting real provider APIs at real $ where a provider call is involved, and paste commands plus full output up to the step that dead-ends, saying in user terms what stopped them, with config, env vars (secrets redacted), version, and the provider's own docs plus a direct call to them when the ask is provider parity
4. Every required box starts genuinely empty, so submitting without typing anything trips GitHub's own required-field error
5. If they file it without that attempt, Agent Shin's feature rubric now reads the request as missing its evidence of the dead-end, and the grace and close comments it posts ask for that same evidence (commands plus output, a screenshot, or a video), not just a description and use case, while listing any evidence they did paste under "What you got right"
6. A maintainer opening the issue sees the call that should work, the exact point it fails, and a ready acceptance criterion

Relevant issues

Linear ticket

Pre-Submission checklist

Please complete all items before asking a LiteLLM maintainer to review your PR

• I have added meaningful tests
• My PR passes all CI/CD checks (e.g., lint, format, unit tests)
• My PR's scope is as isolated as possible; it only solves 1 specific problem
• I have received a Greptile Confidence Score of at least 4/5 before requesting a maintainer review (Greptile reviews automatically once the PR is opened; only comment @greptileai to re-request a review after pushing changes)

Tests: tests/test_litellm/test_github_triage_with_llm.py pins the new feature-request rubric clause in the triage prompt, the dead-end-evidence ask in the bot's close and grace recovery comments, and the "What you got right" credit for evidence a feature requester did provide; the form and the CLAUDE.md line have no runtime code to test

Delays in PR merge?

If you're seeing a delay in your PR being merged, ping the LiteLLM Team on Slack (#pr-review).

Screenshots / Proof of Fix

No live proxy or LLM call applies here, nothing in the request path changed. The form parses, carries the new fields in order, and leaves no required textarea prefilled, at 812abcc:

$ uv run --no-sync python -c "
import yaml
d=yaml.safe_load(open('.github/ISSUE_TEMPLATE/feature_request.yml'))
print([b.get('id') for b in d['body']])
print('prefilled:',[b['id'] for b in d['body'] if b['type']=='textarea' and 'value' in b['attributes']])
"
[None, 'duplicate-check', 'the-feature', 'user-flow', 'how-far-you-got', 'component', 'hiring-interest', 'contact']
prefilled: []

$ uv run --no-sync pytest tests/test_litellm/test_github_triage_with_llm.py -q
121 passed in 0.25s

GitHub only renders an issue form from the default branch, so the visual proof is the form at https://github.com/BerriAI/litellm/issues/new?template=feature_request.yml once this lands. Happy to post a screenshot of it then

Type

📖 Documentation

Caveats (if any)

• Dropped "Motivation, pitch"; the "Before this feature" list carries it
• Asks OSS requesters for a live-proxy run, some will skip it
• Guidance now renders as help text, not editable comments

Final Attestation

• The tests check the right things, including the edge cases, and regressions in the respective real-world customer use-cases are not possible after this PR

Link to Devin session: https://app.devin.ai/sessions/1482580bad6549fda20e10dfeba8400a
Requested by: @mateo-berri

---

Note

^{Cursor Bugbot is generating a summary for commit 812abcc. Configure here.}

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring

[greptile-apps]

Greptile Summary

The PR replaces abstract feature-request prompts with required user-flow and dead-end-evidence fields, then aligns Agent Shin’s triage rubric and recovery comments with those requirements.

• Moves all textarea scaffolding into descriptions and placeholders so required fields begin empty.
• Adds a dedicated triage verdict field for feature-request dead-end evidence.
• Updates close and grace-period guidance and adds regression coverage for the new rubric.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
.github/ISSUE_TEMPLATE/feature_request.yml	Replaces the motivation field with required user-flow and attempted-run fields whose guidance is not submitted as prefilled content; the previously reported required-field bypass is resolved.
.github/scripts/triage_with_llm.py	Requires and credits end-to-end dead-end evidence for feature requests and updates recovery comments consistently.
tests/test_litellm/test_github_triage_with_llm.py	Adds focused assertions covering the new triage verdict field and feature-request recovery guidance.
CLAUDE.md	Extends repository guidance to require reading the feature-request template before filing requests.

_{Reviews (6): Last reviewed commit: "fix(triage): track and credit feature de..." | Re-trigger Greptile}

[devin-ai-integration]

@greptileai

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[devin-ai-integration]

@greptileai

[mateo-berri]

@greptileai

[mateo-berri]

bugbot run

[mateo-berri]

@greptileai

[mateo-berri]

bugbot run

[mateo-berri]

@greptileai

[mateo-berri]

bugbot run

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

^{Reviewed by Cursor Bugbot for commit 812abcc. Configure here.}