Skip to content
View benjaminstelzer's full-sized avatar

Block or report benjaminstelzer

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
BenjaminStelzer/README.md

Benjamin Stelzer

I build practical Agent Skills and developer tools for Codex, Claude Code, and AI-assisted software work.

Why this work?

I use AI agents for real projects, and I keep running into the same problems. An agent adds structure before understanding the existing code, rewrites a sentence and changes its meaning, or reports a result it has not checked.

My Skills put instructions around those failures. They should help the agent finish the actual task without making every small change a process of its own.

Every Skill started with a gap in my workflow or a specific problem I needed to solve.

How I work

I start with the result I need and the project that already exists. Its code, terminology and decisions matter more than a new abstraction or a cleaner description of something that is still wrong.

I use the Skills in real projects and have complete task histories analyzed alongside the results. Those analyses help identify where instructions fail, where they leave too much room for interpretation and where repeated searches, unnecessary checks or oversized output waste context.

I combine targeted simulation runs with several optimization workflows, including SkillOpt. This is an iterative process over months: problems from real projects feed into corrections, the affected cases are tested, and the Skills are refined again as new problems emerge. An optimization proposal is something to test, not a reason to keep the change.

Each Skill has a specific job and works independently. I combine them when the task needs more than one of those jobs.

Scoville Family

Install the complete Scoville Suite, including Workflow Beta, or choose individual Skills below to match your needs. Workflow is included only in the suite.

  • Scoville Code owns engineering scope, implementation, risk, and validation.
  • Scoville Plan owns durable Plans, Work Items, Decisions, and lifecycle state.
  • Scoville Scribe owns wording, terminology, factual meaning, and source fidelity.
  • Scoville UI owns framework-aligned implementation, interface mechanics, accessibility, and rendered evidence, with a standalone design fallback.
  • Scoville WordPress UI Backend owns plugin-owned WordPress admin interfaces, platform components, spacing, accessibility and internationalization.
  • Scoville Design owns visual definition, art direction, design systems, critique and repair.
  • Scoville Handoff transfers active work to another agent or session.
  • Scoville Research turns web, GitHub, and scholarly evidence into a decision-ready, claim-traceable result.
  • Scoville Brainstorm explores materially different mechanisms before selection.

Ask Family

Install the complete Ask Suite for Codex or choose individual second-opinion Skills below to match your needs.

  • Ask Claude for Codex gives Codex a read-only second opinion from Claude Code with selectable model, reasoning effort, budget, and persistent conversations.
  • Ask Claude and SOL for Codex runs Claude Code and a separate SOL session in parallel, returns both read-only opinions together, and preserves both conversations for follow-up questions.
  • Ask Claude and Astra for Codex pairs Claude Code with a fresh Astra project task and keeps both conversations available for follow-up questions.
  • Ask Astra for Review for Codex sends questions and review requests to a fresh Astra project task, returns its answer directly, and keeps the task open for follow-up questions. The adviser asks before archiving it.
  • Ask SOL for Review for Codex sends a read-only review request to a fresh SOL project task.

How it fits together

Start with the part that solves your problem. Code covers engineering work, Scribe covers wording and meaning, and Plan keeps longer work recoverable. The other repositories explain their own scope, setup and development.

If you are building Skills yourself, the instructions and development accounts show the choices behind mine. They are projects I keep revising through use.

Pinned Loading

  1. scoville-code-anti-ai-slop scoville-code-anti-ai-slop Public

    Keeps the engineering result visible through the process around it.

    8 1

  2. scoville-handoff scoville-handoff Public

    Transfers the work state, not the whole conversation.

    2

  3. scoville-scribe-anti-ai-slop scoville-scribe-anti-ai-slop Public

    Better writing without quietly changing the facts.

    3

  4. scoville-ui-anti-ai-slop scoville-ui-anti-ai-slop Public

    Agent Skill for framework-aligned UI implementation, accessibility, interaction, and rendered evidence, with optional Scoville Design coordination.

    2

  5. scoville-plan scoville-plan Public

    Keeps project direction recoverable without turning planning into the project.

    Python 3

  6. scoville-brainstorm scoville-brainstorm Public

    Different solution mechanisms before one answer wins too early.

    4