Refresh API compatibility baselines (CompatibilitySuppressions.xml) under Source/. Deletes existing suppression files, regenerates them via `dotnet pack -p:ApiCompatGenerateSuppressionFile=true`, then reviews the diff and flags any non-`LinqToDB.Internal.*` API changes for explicit user approval.
Audit the linq2db `.claude/` instruction corpus + `CLAUDE.md` + `.claude/scripts/` for duplicated rules, dead references, terminology drift, retired-path mentions, SKILL template gaps, `linq2db.slnx` mismatches, and auto-memory entries that would be better as project-level rules. Reports findings in severity order and offers per-finding patches after explicit user confirmation. Read-only until confirmation.
One-stop dispatcher for `.claude/` maintenance work on linq2db. Surveys staleness signals across the project (`.claude/` audit lag, slnx sync, Knowledge Base cursors, KB issue queue, API baselines drift, permission-allowlist lag) and renders a single table the user can pick from instead of remembering which periodic skill to run when. Each row hands off to the actual maintenance skill that performs the work — `chores` itself runs no maintenance, just routes. Use when the user says "chores", "/chores", "what needs maintenance", or "is anything overdue".
Create a new issue on linq2db/linq2db covering Task/Bug/Feature work types. Verifies claims against the current codebase, drafts a structured body with file:line references, checks for duplicates, proposes issue type + up to 3 labels + milestone, and posts only after explicit user confirmation. Sets the GitHub issue type via REST PATCH after creation.
Free-form Q&A grounded in the linq2db knowledge base. Spawns the kb-research subagent with a curated KB-doc shortlist; the agent answers from KB only and returns a synthesized answer plus citations. Optionally fetches full GitHub issue/PR bodies on demand. Read-only; never modifies the repo or the KB.
Build the persistent knowledge base under .claude/knowledge-base/ from scratch. Phased and resumable — each step is sized to a single session and gates on coverage. Re-running picks up at the first incomplete step. Spawns kb-architect / kb-historian / kb-github-curator / kb-issue-detector as needed; never invents content directly.
Query and act on the detected-issues store under .claude/knowledge-base/detected-issues/. Filter via the shared selection grammar (random N, by area, severity, category, source, file, status). Per-result actions include showing detail, creating a GitHub issue (delegates to /create-issue), driving /fix-issue, marking wontfix / duplicate / dismissed / triaged.
Incremental update for the knowledge base under .claude/knowledge-base/. Reads cursors, fetches deltas (commits, GH issues/PRs/discussions, wiki, code), re-runs the relevant indexer agents scoped to the deltas, and runs a random-sample citation audit. Atomically advances cursors. Interruptible at every cursor boundary.