florianbuetow

Run this workflow from Codex. Behavioral guardrails for Claude Code agents. Nine battle-tested rules (no-guessing, no-stalling, no-preference-asking, no-false-completion, no-skipping, no-dismissing, no-cosmetic, no-caveats, no-flagging) enforced via a Stop hook bash script. Includes analyze, install, test, and update skills for data-driven tuning.

Make motion-graphics videos with Codex and plain JavaScript, no video generator: every frame is a pure function of time, Playwright renders it in headless Chromium, FFmpeg stitches the frames into an MP4, and a contact-sheet check loop catches defects before delivery. setup builds the engine and proves it with a 2-second test render; make runs intake, story beats and one storyboard still per beat for approval before writing the full scene code, then checks and renders. Subcommands: setup, make.

Comprehensive application security toolbox for Codex. 62 skills across 8 frameworks (OWASP Top 10, OWASP API Top 10, STRIDE, PASTA, LINDDUN, MITRE ATT&CK, SANS/CWE Top 25, DREAD), red team simulation with 6 attacker personas plus consolidator, and interactive security education.

Generate arc42 architecture documentation from a codebase. Reads the repo into a structured evidence base, then writes the 12 arc42 sections (Mermaid diagrams where the code gives high-confidence structure, explicit GAP flags where human input is needed — never fabricated). Knowledge base derived from the arc42 framework under CC BY-SA 4.0.

Software architecture quality assessment through structural smell detection, quantitative metrics analysis, antipattern identification, dependency structure evaluation, risk/trade-off analysis, and technical debt measurement. Complements SOLID and beyond-SOLID principle checks with architecture-level health diagnostics.

Critique, improve, and create Codex skills against one standard: a skill is a correction you got tired of making. Target the step that fails, use leading words that recruit training priors instead of coined terms, keep skills small and user-invoked, grow them only from observed corrections, and delete every instruction that does no work. critique reports findings with fixes and edits nothing; improve prunes no-ops, swaps explanation for leading words, and folds recurring corrections from session history in as one-sentence fixes; create mines session history for repeated corrections, checks overlap and default behavior before writing, and drafts a minimal skill. Subcommands: critique, improve, create.

Analyze code and architecture for violations of ten system-level software design principles: Separation of Concerns, Single Responsibility (system-level), DRY, Law of Demeter, Loose Coupling / High Cohesion, Evolvability, Resilience, KISS, Principle of Least Surprise, and YAGNI

Run this workflow from Codex. Keep the Anthropic prompt cache warm during Claude Code sessions — especially during peak hours when usage limits are tighter. Detects your cache TTL (5-min default or 1-hour extended) and schedules pings accordingly, reducing token costs by up to 90% on cached context.

Generate and maintain CHANGELOG.md files from git commit history using Keep a Changelog format with semantic versioning. Creates new changelogs or appends updates to existing ones.

Run this workflow from Codex. Generate and maintain .claudeignore files for Claude Code projects. Analyzes repository structure to identify directories that waste context tokens (caches, build artifacts, large non-source directories) and produces a categorized .claudeignore file.

Analyze source code against Robert C. Martin's Clean Code across nine dimensions: Meaningful Names, Functions, Comments, Formatting, Objects & Data Structures, Error Handling, Boundaries, Unit Tests (F.I.R.S.T.), and Concurrency. Reports structured, actionable findings rated HIGH / MEDIUM / LOW, each with a code location, a verified book reference, the issue, and a concrete refactoring step. A book reference is quoted verbatim from the bundled Chapter 17 registry - a real heuristic code (G30, N5, C3) where the catalog defines one, otherwise a canonical chapter rule (Ch.7: Don't Ignore Caught Errors) - never a synthesised identifier. Findings route by the smallest refactoring edit that resolves them; class, system and architecture concerns are out of scope. Self-contained: the skill and its references bundle everything it uses.

Scout a codebase to identify high-value work, plan the fixes (point and architectural), and produce a ranked map from codebase to tasks -- each tagged with a recommended model -- without ever implementing the fix. Cheap-model agents sweep and score every file by Impact x Opportunity (1-5); the router then runs meta-level pattern analysis, plans point and architectural fixes, and recommends a model per task. Team: target guard, skip gate, formula, profile, parallel scorers, ranker, pattern-grouper (meta-level analysis + architectural fix planning), task-writer (point-fix planning), model-recommender. Objectives: bugs, security, tech debt, dead code, performance, conversion.

Communication-style toolkit for Codex. The tldr skill switches Codex into a military-style BLUF (Bottom Line Up Front) mode: extreme brevity, conclusion first, bullets over prose, runnable commands in code blocks, and ruthless cutting of filler, preambles, and hand-holding. Fires on "tldr", "short mode", or "be brief".

An autonomous research pipeline that uses the Hugging Face API to find, analyze, and synthesize AI research into a technical report. Use this for deep dives into technical AI topics.

Diagramming toolkit for Codex. A router skill dispatches diagram requests to format-specific skills: ascii-art (diagrams drawn with extended ASCII box-drawing characters from CP437 — single ┌─┐, double ╔═╗, shading ░▒▓█ — never + - | /, capped at 80 chars with strict row/column alignment), mermaid (flowcharts, sequence, and other Mermaid diagrams), and wardley (Wardley maps in the WTG2 / wardleyToGo DSL, emitted as .wtg2 for rendering to SVG via wtg2svg). Extensible — add a new skill per diagram format and a row to the router table.

Reverse-engineer distributed system tradeoffs from code, configuration, and architecture artifacts. Analyzes six axes: Consistency & Availability, Latency & Throughput, Data Distribution, Transaction Boundaries, Resilience & Failure Isolation, and Observability, Security & Cost.

Run this workflow from Codex. Reverse the built-in Claude Code limitations discovered in the source code leak. Installs production-grade agent directives that override verification gates, context decay, brevity mandates, and silent truncation — the same fixes Anthropic uses internally.

Create and maintain .gitignore files from a curated, conservative template library. Adds user-specified entries by default, or composes OS-specific (macOS, Windows, Linux) and language-specific (Node.js, Python, Java, Kotlin, C, Go, Elixir) sections without hiding lockfiles, manifests, or other legitimately versioned files.

Helper skill for the guard CLI (https://github.com/florianbuetow/guard), a file-permission tool that protects files from accidental modification by AI agents via immutable flags and root ownership. Composes guard commands for five intents — init, create-collection, remove-collection, clear-all, info — running registry operations directly and printing sudo guard commands for operations that need root.

Session handoff and continuation for Codex. Compact conversations into structured handoff documents for task continuity, then find and resume prior handoffs with full context recovery.

Intent-based task resolution. When stuck, facing ambiguity, or about to ask a clarifying question, first recover the task's underlying intent from available context - the latest prompt, earlier prompts, project goals, plans, and prior corrections - then answer low-risk questions from that intent and make intent-aligned decisions. Escalate to the user only when the ambiguity is material and cannot be resolved from context; and when neither the intent nor a safe default can be inferred at all, an escape hatch stops and names the gap rather than fabricating an answer.

Scan a software repository for ISO 27001:2022 compliance of its software development controls (Annex A 8.4, 8.25–8.33). Produces a per-control compliance gap report covering source code access, secure SDLC, application security requirements, secure architecture, secure coding, security testing, outsourced development, environment separation, change management, and test data.

Analyze code and architecture for unnecessary complexity, over-abstraction, and redundancy. Reports findings with severity ratings and concrete simplification suggestions without modifying code.

Run this workflow from Codex. Analyze Claude Code session logs to generate per-project time and message reports. Breaks down usage by project and branch, groups branches under their parent project, and produces monthly + yearly markdown reports.

Project onboarding skill that orients AI agents by gathering context from project instructions, git state, issue trackers, and build systems to produce a concise status briefing with suggested next steps.

Produces evidence-based model routing plans: maps each task unit to the best-suited AI model (Codex, GPT, Gemini, open-weight) using a 30-category task taxonomy, decision rules, and per-unit rationale, context briefing, and failure-mode warnings.

Analyze and restructure how a repository progressively discloses its documentation through root configuration files (README.md, AGENTS.md, CLAUDE.md, GEMINI.md, USER.md, TOOLS.md, BOOTSTRAP.md, DESIGN.md, NOTICE.md). Maps reference graphs, detects orphaned documents, and generates thematic book-style indexes.

Build and review software engineering résumés from an evidence chain, not a template, in the manner of Richard Feynman and W. Edwards Deming. Applies the three-reader standard (parser, recruiter, engineering interviewer): every claim carries context, action, constraint, result, and ownership; every number is asked by what method; the person's contribution is separated from the system's. review judges a résumé in two phases (up to eight ranked questions, then re-score on the answers into a prediction and a fix list) and never emits an ATS score. build harvests accomplishments one at a time into a win log with provenance, writes five-part bullets, assembles a few coherent versions per role family, runs the two pre-apply tests, and sets up tracking. Subcommands: review, build.

Run this workflow from Codex. Analyze Claude Code session logs to identify what went well, what didn't, and how to improve developer-AI collaboration. Suggests new skills, subagents, slash commands, hooks, and workflow optimizations based on actual usage patterns.

Run this workflow from Codex. Export Claude Code session logs as standard LLM conversation JSON and human-readable TXT. Compact sessions into resumable context files. Recap recent sessions for quick context recovery. Identify current session, export individual sessions, or batch-export all sessions for a project.

Analyze code for violations of the SOLID principles of object-oriented design (Single Responsibility, Open/Closed, Liskov Substitution, Interface Segregation, Dependency Inversion)

Specification-driven development workflow skill. Orchestrates a spec-first discipline: frame a one-page Product Spec (problem, falsifiable bet, success criteria, evaluation) that replaces the PRD, then write behavioral specifications, derive test scenarios, plan implementation, and verify alignment across all artifacts and code. Acts as a workflow navigator with advisory quality gates between phases.

An expert-guided skill for creating layered software specification documents for greenfield projects. Produces Vision, Business Requirements, Software Requirements, Architecture & Design, and Behavioral Spec & Test Verification documents.

Run this workflow from Codex. Stop hooks that terminate a Claude Code session when the agent's final message contains a kill phrase. Two scripts: single-kill (ends Claude) and double-kill (ends Claude plus the terminal shell that launched it), each bound to its own phrase. Installs at local (project) or global (user) scope; scripts require LOCAL or GLOBAL arg and exit 1 if missing. Contains-match (robust to model preambles), case-sensitivity configurable, and a kill path that walks the process tree to the owning claude (and, for double-kill, the top-most shell ancestor) and never signals pid <= 2. Install, remove, update, info, and whendone skills via a router; composes safely with existing Stop hooks.

Run this workflow from Codex. Analyze Claude Code session token usage to flag waste risks and cost optimization opportunities. Reviews per-message token data including cache efficiency, conversation sprawl, model selection, and estimated costs across projects.