Mind-Vault

Alias for /plan — brainstorming is the interactive front-end of the plan skill when input is thin; both commands invoke the same skill

Create well-structured pull request with AI-generated title and description

Show comprehensive git status and working directory state

Load mind-vault behavioural rules (RULE_*.md) — works on both the plugin and symlink channels

Multi-engine review-fix-rerun loop (Cursor Bugbot + GitHub Copilot + Claude Code Review, or any subset) with bounded-autonomy policy

Use this agent for cross-cutting structural design and review — multi-app refactors, dependency/coupling decisions, abstraction boundaries, and blast-radius analysis before feature code is written. It is the reviewer of a drafted plan in /plan, and the author of cross-cutting refactors in /work. Examples: <example> Context: A plan touches auth, billing, and kb apps through a shared base class. user: "Refactor the permission layer across auth, billing, and kb." assistant: "I'll use the architect agent to map the dependency surface and design the shared abstraction before any app-level edits." <commentary> Spans 3 apps with a shared base class — cross-cutting, so route to architect rather than a single-domain implementer. </commentary> </example> <example> Context: A drafted plan needs an independent structural read before execution. user: "Review this plan for coupling and genericity issues." assistant: "I'll use the architect agent to run its abstraction / coupling / boundary / scaling passes over the plan." <commentary> Plan review (not authoring) is the architect's reviewer mode in /plan. </commentary> </example>

Use this agent for Django server-side implementation — models, migrations, signals, DRF viewsets/serializers, Channels, Celery tasks, and ORM optimization (select_related / prefetch_related, killing N+1s, service-layer extraction). Examples: <example> Context: A feature needs a new API surface the frontend will consume. user: "Add a billing_summary API endpoint." assistant: "I'll use the backend agent to add the DRF viewset, serializer, and route with the query optimized up front." <commentary> Models/views/DRF is backend's domain. </commentary> </example> <example> Context: A list view is issuing a query per row. user: "This admin page is slow — looks like an N+1 on the orders list." assistant: "I'll use the backend agent to trace the queryset and add the right prefetch_related/select_related." <commentary> ORM efficiency and N+1 elimination are core backend responsibilities. </commentary> </example>

Use this agent as a relentless pre-commit / pre-push code reviewer — a local Bugbot/Copilot replacement specialized in Django, PostgreSQL multi-tenancy, and HTMX/Alpine patterns. It reviews a diff and reports findings; it never writes or edits files. It has no Write/Edit tools; `Bash`/`Grep` are granted only to inspect (e.g. `git diff HEAD`), so read-only here is a behavioral constraint, not a tool-enforced sandbox. Reach for it when no external review bot is wired up, or before opening a PR. Examples: <example> Context: Work is done on a feature branch and the user wants a gate before pushing. user: "Review my changes before I push." assistant: "I'll use the curator agent to run a multi-pass review over the local diff and report findings by severity." <commentary> Pre-push local review with no external bot is exactly curator's role. </commentary> </example> <example> Context: A PR introduces a ViewSet and the user wants a pattern-enforcement pass. user: "Check this viewset for our multi-tenant and filterset conventions." assistant: "I'll use the curator agent to verify tenant scoping and that filterset_fields excludes removed model fields." <commentary> Pattern/convention enforcement on a diff routes to curator. </commentary> </example>

Use this agent for infrastructure and operations work — Docker / docker compose, nginx & Traefik, systemd, CI/CD pipelines, env/config wiring, and zero-downtime deploy concerns. Assumes failure, enforces idempotency and container parity. Examples: <example> Context: A new background worker needs to run in the stack. user: "Dockerise the new Celery queue." assistant: "I'll use the devops agent to add the compose service, entrypoint, and healthcheck with prod parity." <commentary> Docker/compose/entrypoint work is devops's domain. </commentary> </example> <example> Context: The project needs CI to run tests on every PR. user: "Wire up a GitHub Actions pipeline that runs the test suite." assistant: "I'll use the devops agent to author the workflow with caching and a fail-fast matrix." <commentary> CI/CD authoring routes to devops. </commentary> </example>

Use this agent for documentation-only work — READMEs, guides, CHANGELOG entries, reference docs — written show-don't-tell with runnable examples, negative examples, and zero marketing fluff. Examples: <example> Context: A shipped feature has no user-facing docs. user: "Document the new review-loop engine selector." assistant: "I'll use the documentation agent to write a concise guide with a worked config example and the common failure mode." <commentary> Pure documentation authoring routes to documentation. </commentary> </example> <example> Context: An existing guide is confusing and assumes too much. user: "This setup doc keeps tripping people up — clarify it." assistant: "I'll use the documentation agent to restructure it for an exhausted reader: prerequisites first, copy-pasteable steps, gotchas called out." <commentary> Clarifying/rewriting docs is documentation's job. </commentary> </example>

Search outside the project (in IDE plans, AI agent workspaces, or temporary storage) and retrieve standalone artefacts, research, or validation logs to bring them inside the project repository.

Route a just-learned lesson to the right destination — project-local solution doc, mind-vault skill/rule/agent/command, or auto-memory. Uses a hybrid narrative-probe + taxonomy-quiz router. Also consumes review-loop output (Cursor Bugbot or GitHub Copilot) as input. Final stage of the mind-vault sprint workflow; the lever that makes each sprint easier than the last.

Triage and batch-merge accumulated Dependabot PRs in multi-requirements-file Python repos (web + lsp + dev + per-workspace) — diff-based duplicate detection across the root vs per-workspace ecosystems, risk-tier batching for git-bisect cleanliness, worktree-isolated verification, live-staging smoke, post-merge forward-sync of remaining branches. TRIGGER when the user says "review dependabot PRs", "merge dependabot batch", "dependabot sweep", "what to do with these dep PRs", "clean up the dep updates", or asks for a roadmap merging multiple `chore(deps)` PRs.

Deploy Docker Compose web applications across any project — branch strategy, change-aware scripts, database backup + rollback safety, screen-session remote execution, Let's Encrypt SSL, health checks, and CI/CD wiring.

Apply Django frontend conventions — HTMX partial responses, Alpine.js state, Bulma components, HTMX modal/formset JS contracts, safe query-string generation, dynamic hx-* attribute handling, and Cotton component primitives — pairing with django backend patterns. Includes hard hazard rules every template edit must respect — multi-line `{# …

1 hook across 1 event