cortex-x

One-shot wrapper for the Sprint-sized integration pattern validated in Sprint 2.44. Drives the full flow - discovery questionnaire → plan artifact → workflow dispatch → empirical phase (if applicable) → triage of R2 findings → doc-regen + commit → status report. Use when N>=5 deliverables, framework-touching, and clear acceptance criteria are possible. Triggers (CZ+EN) "/cortex-sprint", "začni sprint", "spustit sprint", "udělej sprint", "sprint kickoff", "start sprint", "ship sprint X", "new sprint", "kickoff sprint". Composes with prompts/cortex-goal.md (turn budget framing), standards/workflows.md (decision tree), standards/documentation.md (doc-regen step), and the multi-phase workflow dispatch pattern from Sprint 2.44 Probe 3. Adds signed r2-verdict.json artifact (Sprint 2.46) replacing [skip-review] as the primary pre-commit gate. v2 schema (Ed25519 + commit_sha + nonce journal + STRICT_SECRET) shipped Sprint 2.46.1.

Senior front-end designer flow — Phase 0 DS bootstrap from repo, Phase 1 intake (image/sketch/vibe inputs), Phase 1.5 plan gate, Phase 2 library + image-gen MCP, Phase 3 parallel worktree exploration with DS-conformance gate, Phase 4 live dev server iteration, Phase 5 handoff via Stitch DESIGN.md + STYLE.md SSOT split. Opt-in Award mode (`--award`) overrides phases with Awwwards-SOTD discipline from reverse-engineering 25 winners. Auto-invokes proactively when user signals UI/UX intent BEFORE writing markup — natural triggers (CZ+EN): "design this/me a landing/dashboard/hero/signup", "navrhni mi web/landing/dashboard/komponentu", "udělej mi web/stránku", "vytvoř hero/landing/sekci/UI", "potřebuju UI/UX/design", "build me a landing/website/homepage", "make it look premium/playful/brutalist", "redesign this", "redesignuj X", or explicit `/designer`. User doesn't need to know the skill exists — cortex offers it. Add `--award` for portfolio/agency/SOTD-targeted work.

One-screen menu of what cortex-x can do right now — list of invokable slash commands (cortex-init, cortex-help, cortex-doctor, audit, designer, start, test-audit) + operator-facing terminal CLIs (cortex-update, cortex-uninstall, cortex-hooks-register, cortex-claude-md-augment, cortex-doctor), a 1-line summary of what each does, and a count of underlying capabilities (action_kinds, profiles, standards). Auto-discovered after install.{sh,ps1} sync. Invoke as "/cortex-help", "what can cortex do?", or when the user asks "co umíš?" / "jaké jsou skills?" / "ukaž mi capabilities". NOTE: namespaced as cortex-help because /help is reserved as a Claude Code built-in slash command. Defers full machine-readable registry to cortex/capabilities.md.

Read-only audit of an EXISTING codebase for module-depth problems that make it hard for AI agents to work in. Surfaces shallow modules (leaky interfaces, many tiny files exporting internals), zero-test modules, and tightly-coupled clusters, then proposes deepening candidates with coupling rationale + a test-boundary recommendation. Produces a prioritized markdown report — NEVER auto-refactors. Grounded in Ousterhout's A Philosophy of Software Design (deep vs shallow modules) + the principle that test/feedback-loop quality sets the ceiling on AI coding quality (SWE-bench Verified). Use after /audit or /test-audit, before a big AI-assisted refactor, or when agents keep producing low-quality changes in one area. Triggers (CZ+EN) "/improve-codebase-architecture", "improve architecture", "deepen modules", "make this codebase agent-friendly", "why does AI write bad code in this repo", "shallow modules audit", "prohlub moduly", "zlepši architekturu", "proč tu AI píše špatný kód", "najdi shallow moduly".

Scaffolds and runs a Ralph Wiggum loop — Geoff Huntley's "stupid simple while-loop" pattern for autonomous overnight coding sessions. Authors PROMPT.md + fix_plan.md + ralph.sh, gates them through a plan review, then runs `while: cat PROMPT.md | claude -p` in a fresh context window per iteration (defeats compaction + context-rot). Useful for long worklists with clear acceptance criteria (test suite passes, fix_plan items checked off), reverse-engineering, spec-driven refactors, "fix all the failing crons overnight" sweeps. Composes with `/cortex-goal` — Ralph is the third option alongside native `/loop` (interval-based) + `/goal` (haiku-verifier). Inherits cortex-x R1+R2 discipline, cost ceilings, and the operator killswitch file. Triggers (EN+CZ) "/ralph-loop", "/ralph", "ralph wiggum", "run ralph", "spusť ralph", "loop until done", "loop dokud nebude hotovo", "overnight loop", "autonomous overnight", "fix all X until tests pass", "oprav všechny X dokud projdou testy", "vibe-clone".

Initialize cortex-x in the current project — interactively pick mode (New / Existing / Framework-only) via AskUserQuestion, write .cortex-bootstrap-pending marker, chain to /start or /audit. RECOMMENDED entry point after install.sh; replaces shell-level cortex-bootstrap CLI. Auto-discovered from ~/.claude/skills/cortex-init/SKILL.md (user-level), works in any project directory. INVOKE PROACTIVELY when user signals installing or using cortex-x — natural-language triggers EN+CZ are load-bearing, don't wait for explicit slash command. Triggers (EN): "let's start", "set up cortex", "init this project", "I installed cortex", "where is cortex installed", "is cortex set up?", "does cortex work?", "/cortex-init". Triggers (CZ): "začni", "nastav cortex", "nainstaloval jsem cortex", "kde je cortex", "je cortex nainstalovaný", "funguje cortex?", "spusť cortex".

Generates bilingual (CZ + EN) UX + conversion copy — runs a structured voice intake, picks the right framework (PAS / AIDA / JTBD / SB7 / Wiebe / Dunford / FAB / BAB), produces hero copy / landing-page sequences / CTAs / microcopy / error messages / onboarding / empty states with hard length budgets, AI-slop blocklist, and CZ-localization linting (tykání-vykání tree, EN→CZ +30% expansion, Smazat/Vymazat/Submit traps, Mozilla CZ typography). Auto-invokes on copy intent — triggers (CZ+EN) "napiš mi text/headline/hero/slogany", "potřebuju copy", "přepiš tlačítka", "lokalizuj UI do CZ", "write copy for", "draft a landing page", "rewrite my hero", "fix my microcopy", or explicit `/copywriter` / `/ux-copywriter`. Grounded in May 2026 research across conversion frameworks, content-design canon (Microsoft / GOV.UK / Apple / Material / Mailchimp / Atlassian / Shopify / Carbon), Podmajersky Voice Chart, and Mozilla CZ Style Guide.

External tool capability adapter for Hyperframes (HTML → video, agent-native). Sprint 3.4 v0 ships the SKILL.md frontmatter contract + license-tier gate; Sprint 3.4 v1 wires the executor + Docker sandbox + first end-to-end render. Apache-2.0 OSS; no per-render fees; deterministic ("same input = identical output"). Operator invokes via "/external-adapter-hyperframes" or natural-language equivalent ("render the landing as a 30-second hero video"). Auto-discovered after install.{sh,ps1} sync.

Plan-first wrapper for Claude Code's native /goal slash command. Produces a structured cortex-disciplined plan (brief · scope · DoD · acceptance criteria mirroring cortex spec-verifier kinds · R1+R2 discipline embedded) that the operator hands to native /goal. Useful for autonomous overnight runs, multi-hour focused sessions, weekly-token-frontload before rate-limit reset, and any task with a clear acceptance frontier and turn budget. Triggers (EN+CZ) "/cortex-goal", "cortex goal", "run long task", "autonomous session", "overnight", "use my unused tokens", "execute this overnight", "weekend run", "vyřeš to autonomně", "dělej dokud nehotovo", "běhej v noci", "dlouhá session", "úkol na celý víkend". Composes with /goal native, not replaces it — cortex authors the plan, Claude Code runs the haiku verifier loop.

Remove cortex-x from this machine — framework, shims, skills, agents. Wraps the bin/cortex-uninstall.cjs CLI. Safe-by-default — preserves $CORTEX_DATA_HOME (~/.cortex/) which holds research/projects/journal/insights (months of work). Triggers (EN+CZ) "/cortex-uninstall", "uninstall cortex", "remove cortex", "smaž cortex", "odinstaluj cortex", "wipe cortex", "i don't want cortex anymore". Refuses to act without explicit operator confirmation. The --purge flag (deletes ~/.cortex/ too) is DESTRUCTIVE and requires a second confirmation step on top of the first. Always runs --dry-run first so the operator sees the exact path list before any deletion happens.

Update the cortex-x install on this machine — fetch latest source, fast-forward if possible, re-run installer. Wraps the bin/cortex-update.cjs CLI. Useful weekly or when CHANGELOG mentions a new version. Triggers (EN+CZ) "/cortex-update", "update cortex", "upgrade cortex", "cortex je staré", "aktualizuj cortex", "what's new in cortex", "is there a cortex update", "máš novou verzi cortexu". Safe by default — refuses on dirty tree, only fast-forwards (never rewrites history), aborts cleanly on network failure. After update finishes, recommends `claude` session restart so Claude Code reloads the refreshed `~/.claude/shared/`.

Health check + drift detection for the cortex-x install on this machine. Runs the bin/cortex-doctor.cjs CLI, parses findings, and offers interactive auto-fix for each non-ok check (Y/n per finding). Useful weekly, after a fresh install, after a migration to a new machine, or when cortex behavior feels off. Triggers (EN+CZ) "/cortex-doctor", "/doctor", "doctor", "healthcheck cortex", "co je rozbité na cortexu", "is cortex healthy", "cortex není v pořádku", "cortex broken". Surfaces hooks-registered status, CLAUDE.md discipline block status, source clone state, data home subdirs, skills + agents discovery. Composes with cortex-hooks-register and cortex-claude-md-augment for one-tap remediation.

Deep 12-dimension audit of an EXISTING codebase via cortex-x existing-project-audit prompt (P0 detect → P1 repo-map → P2 4-agent audit → P3 human gate → P4 research → P5 synthesis → P6 ADR backfill opt-in). Saves AUDIT.md, recommendations.md, repo-map.md, and optional retroactive ADRs to cortex/ in the project. Auto-primed by SessionStart hook when .cortex-bootstrap-pending marker is mode=existing. Triggers (CZ+EN): "/audit", "audit my project", "audituj projekt", "co je v tomhle projektu", "prozkoumej kódbázi", "review this repo", "what does this codebase do".

Bootstrap a NEW project from an empty folder via cortex-x five-phase onboarding (Discover → Research → Architect → Scaffold → Adapt). Saves artifacts to cortex/ at each phase. Auto-primed by SessionStart hook when .cortex-bootstrap-pending marker is present. Triggers (CZ+EN): "/start", "let's start a new project", "začni nový projekt", "scaffold this", "naskafolduj mi projekt", "založ nový repo", "bootstrap new app".

Deep QA-focused audit of an EXISTING codebase via cortex-x qa-retrofit prompt (P0 detect → P1 inventory → P2 4-agent ISO 25010:2023 + cortex extras → P3 human gate → P4 QA-specific research → P5 synthesis → P6 sample-test seeding opt-in). Produces senior-QA-consultant-grade testing strategy + prioritized gap backlog. Pairs with /audit (general 12-dim) but with a testing lens. Triggers (CZ+EN): "/test-audit", "test strategy review", "ohodnoť testy", "jak na tom jsou testy", "QA audit", "test coverage assessment", "review my tests".