claude-code - 💡(How to fix) Fix [FEATURE] Ship a first-party feedback skill for filing issues from inside Claude Code [1 participants]

Preflight Checklist

• I have searched existing requests and this feature hasn't been requested yet
• This is a single feature request (not multiple features)

Problem Statement

Filing well-formed issues against anthropics/claude-code from inside an active Claude Code session is higher-friction than it should be. The in-repo issue forms (bug_report.yml, feature_request.yml, model_behavior.yml, docs.yml) have many required fields, strict dropdown options, and change over time. Ad-hoc issue creation from a session tends to produce malformed reports that bounce or need re-filing, and the built-in /bug flow only covers a subset of the feedback surface (no feature requests, docs issues, or model-behavior reports).

Community plugins can paper over this, but they go stale the moment upstream templates change, aren't discoverable to most users, and have no sanctioned status.

Proposed Solution

Ship a first-party feedback skill (or skill bundle) maintained by Anthropic, bundled with Claude Code or available via the official plugin marketplace.

Here's a functional skill I built that covers this surface: https://github.com/danielrosehill/Claude-Code-Feedback-Plugin — I'm offering it as a starting point / reference implementation, but the point of this request is that something like this should be an Anthropic-maintained agent skill rather than a community plugin. Happy to transfer, relicense, or have it reimplemented; attribution isn't the ask.

Minimum viable shape:

• One skill per issue-form type under .github/ISSUE_TEMPLATE/ (bug, feature request, model behavior, docs — plus anything Anthropic adds later).
• Self-healing: fetch the live YAML from main on each run rather than baking field lists into the skill. Cache for offline fallback and drift detection.
• Interview-style gathering that respects required/validations, renders dropdowns verbatim, and previews the final body before gh issue create.
• Invoked naturally ("report a bug", "file a feature request", "the model refused to…") so users don't need to remember a slash command.

Because Anthropic controls both the issue forms and the skill, self-healing becomes near-trivial to guarantee — template changes can ship alongside skill updates.

Alternative Solutions

• Status quo: users open the GitHub web UI and fill the form manually — context-switch cost suppresses feedback volume.
• The existing built-in /bug command: bugs only, and has had its own issues (e.g. #35845 where user skills intercepted it). Doesn't cover feature requests, model behavior, or docs.
• Third-party plugins: work for users who find and install them, but fragmentation means most users never encounter one, and maintenance falls on individuals.

Priority

Low - Nice to have

Feature Category

Developer tools/SDK

Use Case Example

1. Mid-session, Claude Code does something surprising — a skill stalls, or the model returns an obviously wrong refactor.
2. User types "file a bug about this" (or "feature request: I wish /resume could…").
3. The first-party feedback skill activates, fetches the current bug_report.yml from main, and walks the user through each field with context from the current session pre-filled where appropriate (CLI version, OS, recent tool calls — with secrets redacted).
4. Skill previews the rendered issue, user confirms, gh issue create submits.
5. User stays in flow; Anthropic receives a correctly-formatted report that slots straight into triage.

Additional Context

Reference implementation: https://github.com/danielrosehill/Claude-Code-Feedback-Plugin — MIT-licensed, four skills (bug / feature-request / model-behavior / docs) plus a refresh-templates skill that detects drift and scaffolds new skills when Anthropic adds new issue-form types.

Related existing requests: #30727 (Official Verified Marketplace for Skills, MCP Servers, and Custom Agents) — this is a narrower first-party-skill ask, not a marketplace proposal.