Skill Claude CodeCodex
Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a…
Skill Claude CodeCodex
Log into the local Carbon ERP dev server with agent-browser using the DEVBYPASSEMAIL bypass. Use before any browser automation that needs an authenticated session (/test, /smoke-test, manual verification). Requires a running dev stack (crbn up). Building block — it leaves the session open for the caller.
Skill Claude CodeCodex
Author, edit, or extend the Carbon documentation site at docs (a built Fumadocs + Next.js app). Use whenever creating or changing reader-facing docs for Carbon: editorial Guide chapters, Reference/entity pages, the docs IA, or "document this feature" requests — anything destined for the docs site. Covers the…
Skill Claude CodeCodex
Pre-commit verification gate — runs Carbon's validation gates in order (generate:types if schema changed, biome, scoped typecheck, scoped tests, build if needed, and /translate to fill missing i18n .po strings when UI/locale files changed), fixes straightforward failures, then commits the specific files with a…
Skill Claude CodeCodex
Autonomous loop conductor — drive a single work item (a bug, a usability tweak, a small feature) through a doer→gate→judge→keep-or-revert→ledger cycle to a gated PR, unattended. Use when asked to "conduct", "loop on", or build/fix one tightly-scoped item with explicit acceptance criteria, and as the inner loop the…
Skill Claude CodeCodex
Create or refresh an AGENTS.md for a Carbon package or ERP module by reading the actual source code — every claim traced to a real function, table, or export. Use when adding a new module/package, when the conductor's freshness audit or /self-review flags a stale AGENTS.md, or on explicit request. Do not write an…
Skill Claude CodeCodex
Runtime-instrumentation debugging for bugs that static reading can't pin down — add temporary unconditional JSONL logging to the real code path, reproduce, read the log, then fix. Use when /root-cause lands at MEDIUM/LOW confidence, when a bug involves runtime state, ordering, caching, streaming, concurrency, or…
Skill Claude CodeCodex
Capture a screenshot plus element snapshot of the current browser page when an error appears during e2e testing, saved to gitignored .ai/scratch/e2e/. Use whenever an agent-browser snapshot shows "Something went wrong", an auth error, a blank page, or an unexpected redirect. Building block — it captures and returns…
Skill Claude CodeCodex
Execute an approved implementation plan from .ai/plans/ task by task, running each task's verification and committing per task via the check-and-commit gate. Use when asked to "execute the plan", "implement the plan", or after /plan approval. Do not use without an approved plan file — run /plan first — and do not use…
Skill Claude CodeCodex
End-to-end feature pipeline that orchestrates /research, /spec-writing, /plan, /execute, /test, and /self-review. At start it picks an autonomy mode (approval-before-each-phase vs fully autonomous) and a phase set (plan + execute are mandatory; research, spec, test, self-review are optional), auto-detected from the…
Skill Claude CodeCodex
End-to-end bug-fix pipeline that diagnoses, implements, and proves the fix. It orchestrates /root-cause, /debugging-difficult-bugs, /test, and /check-and-commit, and performs the fix itself — the minimal code change from the proven cause plus a mandatory red→green regression test and scoped validation gates. At start…
Skill Claude CodeCodex
Interview the user relentlessly, one question at a time, to stress-test a plan, spec, or design until every open decision is resolved — recommended answer per question, answers cross-checked against the codebase and the @carbon/glossary domain terms, resolutions written back to the artifact as they land. Use when the…
Skill Claude CodeCodex
Survey the Carbon codebase as a senior advisor and produce prioritized, self-contained implementation plans for OTHER models/agents to execute. Strictly read-only on source code — never implements, fixes, or refactors anything itself. Use when asked to audit Carbon, find improvement opportunities (bugs, security…
Skill Claude CodeCodex
Inngest platform reference for Carbon's background jobs — durable functions, steps, events, flow control, error handling, and local dev, all in the v3 syntax Carbon actually runs (inngest ^3.52 in @carbon/jobs). Use when writing or reviewing Inngest functions, choosing flow-control…
Skill Claude CodeCodex
Design engineering principles for making interfaces feel polished. Use when building UI components, reviewing frontend code, implementing animations, hover states, shadows, borders, typography, micro-interactions, enter/exit animations, or any visual detail work. Triggers on UI polish, design details, "make it feel…
Skill Claude CodeCodex
Turn a finalized spec into a step-by-step implementation plan at .ai/plans/{YYYY-MM-DD}-{slug}.md, where every task has exact file paths, exact commands, and a verification with expected output. Use when asked to "plan the implementation", "create a plan", or after a spec's open questions are resolved. Do not use…
Skill Claude CodeCodex
Build a self-contained HTML review aid at .pr-review/{branch}.html that teaches a reviewer what a PR changes and why — problem, system context, before/after flow, focused diffs, verification evidence, takeaway. Use when asked to explain a PR, help reviewers understand a complex change, or produce a review page for a…
Skill Claude CodeCodex
Split a large, messy, or hard-to-review pull request into smaller reviewable PRs (stacked or parallel) without losing any of the original work — snapshot the branch, extract changes non-interactively, verify each PR independently, and track drift as review feedback lands. Use when a PR is too big to review, mixes…
Skill Claude CodeCodex
Survey how best-in-class ERPs and point solutions implement a feature, and produce a findings file at .ai/research/{slug}.md. Use when designing a new feature, evaluating approaches, or answering "how do competitors / SAP / the industry do X". Required before designing any ERP-domain feature (accounting, costing, tax…
Skill Claude CodeCodex
Read-only root-cause analysis for any bug, test failure, or unexpected behavior — before proposing or writing any fix. Produces a brief with the root cause, files to change, approach, and risks. No edits, no commits, no state-changing commands. Use before /fix or /conductor whenever the cause isn't already proven. If…
Skill Claude CodeCodex
Guide for writing idiomatic Rust code based on Apollo GraphQL's best practices handbook. Use this skill when: (1) writing new Rust code or functions, (2) reviewing or refactoring existing Rust code, (3) deciding between borrowing vs cloning or ownership patterns, (4) implementing error handling with Result types, (5)…
Skill Claude CodeCodex
Rust reference for Carbon's cargo workspace (crates/ + apps/assembler) — tokio discipline for CPU-heavy C++ FFI, ownership/interior-mutability choices, rayon isolation, error conventions, performance and profiling, all grounded in measured results from this codebase. Use when writing or reviewing Rust in this repo…
Skill Claude CodeCodex
Critically review your own branch work before or just after opening the PR, producing Must fix / Risks / Suggested improvements plus a docs-freshness check. Use when finishing a branch, before opening or merging a PR, or to sanity-check a diff against main. Supports an opt-in strict "thermo-nuclear" / "nuclear review"…
Skill Claude CodeCodex
Quick e2e smoke test of the local Carbon ERP dev server — logs in via /auth, then loads each core module and verifies it renders without errors. Use after booting a stack, after wide-reaching changes, or when asked to "smoke test" the app. For feature-specific testing use /test instead.