dte-skills

Make a Rails app installable and offline-capable — web app manifest, service worker, install prompt, offline/cache states, and (optionally) push — built on Rails 8's native PWA support, not a hand-rolled service worker. Composes Rails 8 native PWA (app/views/pwa/manifest.json.erb + service-worker.js + the routes), the ui.sh family + frontend-design for install/offline UI, ie-experience-reviewer for the offline UX, and a Chrome DevTools Application-panel installability check. Use when the user says "make this a PWA", "make it installable", "add a web app manifest / service worker", "work offline", "add to home screen", or "installable web app". Loops over a list of apps/surfaces. Not for general front-end work (use dte-ux) or native mobile (Hotwire Native is a different toolkit).

Architectural review of a Rails app, a subsystem, or a vertical slice, producing a written findings doc. Composes Augment context retrieval, layered-rails analyzers, intent-engineering audit, majestic Rails reviewers, and a ponytail over-engineering pass into one severity-ranked report. Use when the user asks to "review the architecture", "do an arc review", "assess the structure/design", "find architectural issues / god objects / fat models / layering problems", or to review a slice/subsystem. Loops over a list of slices when given several. Not for line-by-line PR/diff review (use dte-deep-reviewer) or for planning fixes (use dte-arc-plan).

Audit a test suite for real value, not just presence — coverage gaps AND test quality (weak assertions, over-mocking, implementation-coupled/brittle tests, missing edge/error/integration paths), plus mutation testing where available to prove tests actually fail when code breaks. Composes SimpleCov, rails-testing (Minitest), the ce-code-review testing persona, and mutant. Use when the user says "audit the tests", "are these tests any good", "check test coverage/quality", "do the tests actually test anything", or after a feature lands. Loops over a list of files/modules. Not for writing new tests (use minitest-coder / rails-testing) — this judges existing ones.

Turn architectural findings, a spec, a bug report, or an "improve area X" request into concrete, validated implementation plan(s) — always built with compound-engineering's ce-plan and validated with intent-engineering's ie-validate-plan. For a massive task it decomposes the work into many small, independently-shippable plans (loop), each ce-planned and validated, and emits a plan index that dte-arc-work then completes. Use when the user says "plan the fixes", "make a plan(s) from these findings", "plan this bug/refactor/feature", "break this big thing into plans", or hands over a dte-arc-review doc. Not for executing plans (use dte-arc-work) or producing findings (use dte-arc-review).

The full cycle for a task, bug, area, or a whole plan index — review, plan, do the work, then verify — with quality gates and progress reporting throughout. Composes dte-arc-review (or a light scan), dte-arc-plan, ce-work, ce-simplify-code/ponytail, and dte-deep-reviewer + dte-test-auditor to verify. Loops over a list of tasks/bugs OR over a dte-arc-plan INDEX worklist to complete many plans (one branch + gate + report each). Use when the user says "fix this", "do the work", "implement this plan", "work through these plans/tasks", or "handle this end to end". Stops for review after each item by default; --ship to auto commit/PR. --quick for a small bug, --deep for a real change. Not for a pure root-cause bug hunt (use dte-debug), a full-stack feature with UI (use dte-feature), an upgrade/data migration (use dte-migrate), or setting up an autonomous batch loop (use dte-loop).

Full-stack new feature spanning front-end AND back-end — plan, implement, and verify across architecture/back-end AND UI/UX/front-end, making sure BOTH are solid. Composes dte-arc-plan (validated plan), dte-arc-work (gated build), dte-ux + the ui.sh family for the front-end, and dte-deep-reviewer + ie-experience-reviewer to verify both layers. Use when the user says "build this feature", "ship this idea end to end", "add <feature> with a UI", or describes something that needs both a model/controller/job AND a screen. Loops over a list of features. Not for a back-end-only change (use dte-arc-work), pure UI (use dte-ux), or just defining the what (use dte-spec).

Plan and execute a safe upgrade or migration — a Rails version bump, gem upgrades, or a schema/data migration — incrementally, reversibly, gated at every step. Composes gemfile-upgrade + gemfile-organize (dependency moves), database-admin + database-optimizer (schema/data), gem-research (is an upgrade safe/maintained), ce-plan/ce-work, and the project gate after each step. Use when the user says "upgrade Rails/this gem", "bump dependencies", "migrate the schema/data", "backfill this column", "zero-downtime migration", or "update our dependencies". Loops over a list of upgrades. Not for an architecture/structure migration (use dte-arc-work from a dte-arc-auditor roadmap) or a code bug (use dte-debug).

Mine session history and recent work to recommend which tools, skills, plugins, or agents to adopt next — and flag overlap with what is already installed. Composes ce-sessions (prior-session synthesis), MemPalace history/knowledge-graph, the overlap logic from the skill-vet design, and light web research for candidates. Use when the user asks "what tools should we add", "scan our history for gaps", "what are we missing", "what would help based on what we do", or "tooling review". Optional loop over time windows or topics. Produces a recommendations doc; it does not install anything. Not for reviewing code or a PR (use dte-deep-reviewer), app architecture (use dte-arc-review), or auditing the dte-* skills themselves (use dte-skill-audit).

Deep multi-lens review of a pull request, a branch diff, or a whole codebase, producing a written findings doc. Runs compound-engineering's cross-model code review, the intent-engineering lenses, majestic's Rails code-review orchestrator, and cubic review/scan, then adversarially cross-checks where they disagree. Use when the user says "review this PR / branch / diff", "deep review", "check this codebase", or hands over PR numbers/branch names. Loops over a list of PRs or branches. Not for architecture-level structure (use dte-arc-review) or for writing/fixing code (use dte-arc-work).

Project-management — turn a goal, spec, or plan index into TRACKED, broken-down, sequenced work (GitHub issues/tickets) with status and dependency order. Composes ce-plan/ce-brainstorm to decompose, dt-create-issue (or gh issue create) to file each ticket with the plain-language What/Why/How body + labels, and produces a tracking index. Use when the user says "track this", "create issues/tickets for this", "break this into tickets", "project-manage this", "file the backlog", "sequence this work", or "what's the status/order". Loops over a list of goals. Not for defining the what (use dte-spec), the implementation plan (use dte-arc-plan), or doing the work (use dte-arc-work).

Audit the dte-* skills themselves against the AGENTS.md quality bar — does each skill compose (not reinvent), loop, degrade gracefully, have sharp non-overlapping triggers, read conventions first, and write a useful doc? Cross-checks trigger overlap across the whole suite and produces a scored report. Read-only — it does not edit skills. Use when the user says "audit the dte skills", "are these skills solid", "skill-audit", "check the suite for overlap/bloat", or after adding several skills. Loops over the whole skills/ directory by default. Not for reviewing application code (use dte-deep-reviewer) or app architecture (use dte-arc-review).

Track a bug to its ROOT CAUSE and fix it once where all callers route — for a failing test, an exception, a wrong result, or flaky behavior. Composes ce-debug (root-cause hypothesis loop), majestic rails-debugger, Augment for tracing call paths, AppSignal for production error context, and a runnable repro before and after the fix. Use when the user says "debug this", "why is this failing/broken", "fix this bug", "this exception", "find the root cause", or "this test is flaky". Loops over a list of bugs. Not for a broad code review (use dte-deep-reviewer) or a planned feature (use dte-feature).

Set up an autonomous, gated loop over a batch of work — turn "I have many plans to implement / reviews to run / bugs to fix" into a worklist doc the loop iterates, with every predictable decision asked UP FRONT, then emit the exact command to trigger it. Each item runs on its own branch, sequentially, gated, with approve→merge criteria you set. Composes the built-in /loop, dte-arc-plan (INDEX worklist), dte-arc-work (gated executor), and dte-deep-reviewer/dte-arc-review for review batches. Use when the user says "loop over these", "run these autonomously", "implement all these plans each on its own branch", "set up a loop", "batch these reviews and merge the approved ones", or "what command do I run to loop this". Not for doing one task end-to-end (use dte-arc-work) or building one plan (use dte-arc-plan).

Performance review of a Rails app, endpoint, query, job, or front-end page — find the real bottlenecks (N+1s, missing indexes, slow queries, fat allocations, slow front-end paint/bundle) with measured evidence, and propose ranked fixes. Composes majestic performance-reviewer + database-optimizer + database-admin, AppSignal production data, chrome-devtools performance traces + Lighthouse, and Augment for context. Use when the user says "it's slow", "find the bottleneck", "perf review", "optimize this query/endpoint/page", "why is this N+1", or "check performance". Loops over a list of endpoints/queries/pages. Not for general architecture (use dte-arc-review) or fixing without measuring (this measures first).

Security review of a Rails app, a PR/diff, or a slice — find real vulnerabilities (authz gaps, injection, mass-assignment, secrets, unsafe deserialization, SSRF, CSRF, privacy/PII leaks) with evidence and ranked fixes. Composes the security-audit (whole-app posture) and security-review (PR-scoped) skills, majestic privacy-reviewer + data-integrity-reviewer, intent-engineering's predictability lens, and Brakeman/bundler-audit if present. Use when the user says "security review", "security sweep", "is this safe", "check for vulnerabilities", "audit auth/authorization", "scan for secrets/PII", or before shipping a sensitive change. Loops over a list of slices/PRs. Not for general correctness review (use dte-deep-reviewer) or architecture (use dte-arc-review).

Turn a fuzzy idea or goal into a right-sized, validated specification (PRD) — the WHAT and WHY, with scope, user stories, acceptance criteria, edge cases, and non-goals — before any planning. Composes ce-brainstorm (explore/scope the idea), ce-ideate, and ie-validate-plan (validate the spec for clarity, simplicity, predictability, UX completeness). Use when the user says "spec this out", "write a PRD", "define the requirements", "what should this do", "scope this feature", or hands a one-line idea that needs shaping. Loops over a list of ideas. Not for the HOW/implementation plan (use dte-arc-plan) or building it (use dte-feature/dte-arc-work).

Designer / UX-UI specialist — review AND produce front-end work: design fidelity, interaction states, look-and-feel, responsiveness, dark mode, and accessibility, ending in a findings doc and/or applied UI. Composes the ui.sh family (ui, ui-design, ui-componentize, ui-make-responsive, ui-add-dark-mode, ui-markup-from-image), frontend-design, and ie-experience-reviewer. Use when the user says "review the UI/UX", "design this screen", "improve the look-and-feel", "check accessibility", "make it responsive", "build this component", "turn this mockup/screenshot into UI", or "does this feel polished". Loops over a list of screens/components. Not for back-end/architecture (use dte-arc-review) or a full front+back feature (use dte-feature).

Full architectural audit of a whole Rails app (or a major subsystem) that produces not just findings but a complete, validated, loop-executable migration roadmap — current-state classification, a proposed target architecture, a phased plan with gates, a risks/decisions doc, and a PROGRESS worklist. This is the "re-architect this app" deliverable (e.g. service-junk-drawer → modular monolith / Packwerk / interactors). Use when the user says "audit the architecture", "do a full architectural audit", "re-architect / restructure the app", "plan a modular-monolith / Packwerk migration", or "assess this app and give me a migration plan". Composes dte-arc-review (current state) + a target-design step + dte-arc-plan (the validated roadmap). Not for a single slice's findings (use dte-arc-review), a PR/diff (use dte-deep-reviewer), or executing the roadmap (use dte-arc-work).