Slop Fix Repo Issues

Start only when the human explicitly invokes $slop-fix-repo-issues to fix repository issues and supplies selection conditions. That invocation and the supplied conditions authorize $bulk to create an item worktree chat for each issue in the fixed list collected for this run, and $slop-fix-issue to handle only that assigned issue through evidence-backed closure or PR creation. Loading, editing, or planning this skill does not start a run.

This is a one-shot batch. Do not register a watch or heartbeat, invoke PR maintenance, or add newly discovered issues after fixing the list. Collect results through issue closure or verified PR creation without waiting for CI, reviews, or merging.

Require selection conditions

Accept the human's issue-selection conditions in natural language or as a GitHub search expression. If conditions are missing or ambiguous, ask a focused question and wait for the answer before searching or assigning issues. A bare skill invocation is not a request to process every open issue. An explicit request for all open issues is a valid selection condition.

Use repo:<owner/repo> is:issue is:open as the base scope. Translate natural-language conditions faithfully and combine them with that scope, preserving the meaning and grouping of supplied search expressions. Add author, label, or other selection filters only when the human specifies them. If conditions conflict with the current repository's open-issue scope, clarify before starting rather than silently changing or discarding them.

Resolve and collect once

1. Read $bulk and $slop-fix-issue, resolving their installed SKILL.md files to absolute paths. PR publication through $slop-fix-issue additionally requires $create-pr. Report unavailable skills instead of substituting another workflow. Let each skill perform its own required preflight and let $slop-fix-issue check and invoke $create-pr when needed; evidence-backed issue closure does not need that publishing dependency.
2. Resolve the current checkout's GitHub host and repository through its remote and authenticated gh access, including the owning repository when already in a worktree. Ask if the repository cannot be resolved uniquely. Resolve the exact query from the required human conditions and base scope. Only when the human selects author:@me, including a natural-language request for issues they authored, record the authenticated login for that host and resolve that qualifier to the recorded author so the collection cannot change scope with authentication identity.
3. Search issues through authenticated gh on the resolved host using the exact query. Fetch all relevant pages and exclude pull requests. Check incomplete_results, the reported total, and pagination completeness; GitHub search exposes at most 1,000 results per query. A failed, partial, capped, or incomplete search is not an empty result. Report the limitation and do not dispatch from it or silently narrow the query.
4. Once collection succeeds, keep one canonical issue URL per verified host/repository/issue-number identity, preserving first-occurrence order. Record the fixed list, original conditions, exact query, recorded author when applicable, and authorization context in this chat. An empty list is a completed no-op: report it and stop without invoking $bulk or creating chats.