ai-config-kit

Route substantive coding, planning, architecture, debugging, review, security, and research across the active coordinator, an independent external model family such as OpenCode/GLM, and optional fast, balanced, or deep Codex peers. Use when adaptive multi-model orchestration is adopted or requested, including deciding whether normal strong effort is enough or max/ultra is justified.

Route into the albrand/ai-config-kit operating framework and load only the relevant doctrine for planning, implementation, debugging, review, security, harness design, quality convergence, repository adoption, journaling, skill learning, token economy, or context acceleration. Use when the user asks to apply, inspect, adopt, or update ai-config-kit; asks for a rigorous agent operating model or harness; or when an installed ai-config-kit specialist skill requires shared framework context.

Perform a whole-project assessment, generate or refresh knowledge docs, derive missing requirements and tickets from internal and external sources, and plan hardening work. Use when the user wants an existing project read deeply and turned into docs, gaps, QA logic, and implementation plans.

Turn business requirements, docs, designs, existing boards, legacy roadmap material, and stakeholder prompts into an evidence-backed roadmap, operating model, and PR-sized ticket ecosystem. Use when the user wants AI to bootstrap, import, reconcile, or create product/project roadmap artifacts for a new or legacy app.

Bootstrap or reconcile the technology side of a project from a stack prompt, existing repos, infrastructure choices, cloud/provider access, local environment needs, CI/CD gates, PR automation, and AI developer workstreams. Use when the user wants a project-agnostic technical framework setup plan or live setup actions.

Turn a product or platform module idea into an evidence-backed delivery plan using AI analysis plus normal repository, documentation, and board inspection. Use when a user provides a module name, capability area, roadmap item, Linear/project-planning request, or migration target and wants Codex or Claude Code to study available docs/repos/boards, ask clarifying questions when needed, define scope, create or update a project description, produce simple phase milestones, PR-sized implementation tickets, resource links, risks, owners, and validation gates for leadership, product, and technical audiences. This skill is intentionally script-free and must not depend on Python, generated scripts, or custom automation.

Review pull requests or diffs with a business-rule-first, high-signal, low-false-positive workflow, mandatory PR review output contract loading, inline code threads, root-cause commentary, practical failure examples, GitHub suggestion blocks when safe, no monolithic review bodies, no validation-transcript boilerplate in PR surfaces, no AI signatures, and a board-backed regression gate. Use when the user asks Codex to review a PR, inspect a diff for merge readiness, prepare review comments, check a branch before merge, evaluate AI-generated code, improve developer-facing PR feedback, or draft PR bodies.

Multi-pass, multi-lens security review that reinforces detection through independent blind finder passes and an adversarial refute pass. Use when reviewing an owned or authorized codebase/change for vulnerabilities with high confidence, or when a change touches auth, data, crypto, external input, dependencies, or build/config files. Static by default; no live target required.

Authorization-gated white-hat penetration testing against an owned or explicitly authorized target. Follows find → validate → fix → regress. Requires the authorization gate before any active testing (crafted requests, exploit runs, scanning, fuzzing). Use for "pentest my app", "test this endpoint for vulnerabilities", or validating an owned finding is truly exploitable.

Find and load the right Codex skill from a large local library when skill descriptions are shortened, hidden, explicit-only, or no visible skill clearly matches. Use before giving up on skill context, after skill or plugin library changes, or when Codex needs smart access to all installed skills without injecting every skill description into the initial context.

UX/product workflow for design-makers and consumers of existing designs. Use for making or revising mockups, screens, layouts, flows, prototypes, tokens, typography, components, design systems, live design previews, signoff, and Figma or board handoff. Also use to consume a prototype, Figma file, screenshots, or UI repo; inventory screens, flows, components, tokens, states, responsive and accessibility behavior; identify open questions and dependencies; and shape PR-sized backlog tickets in Jira, Linear, or another tracker. Detect and announce design-maker or backlog-shaping mode, keep one design source of truth, verify tool access, and approval-gate external writes.

Gate and use optional repository context accelerators such as Graphify-compatible graphs/context maps, OpenWiki-compatible generated agent wikis, symbol indexes, or code-review graphs. Use when the user or repo opts into one of these tools, asks for token-saving context strategy, scope mapping, blast-radius analysis, business-rule discovery, accelerator adoption guidance, or repo-local accelerator skill and instruction updates.