The Problem

LLM-generated infographics look polished but are often technically wrong — mislabeled components, invented values, incorrect relationships. For educational material that gets scrutinized by engineers and students, that failure mode is disqualifying. This repo packages a Codex skill that constrains an agent to produce technically accurate diagrams rather than plausible-looking ones.

What This Does

It is a Codex skill definition, not an application. Four declarative files tell an agent how to research, structure, and render an infographic. SKILL.md is the instruction set the agent follows; references/transistor-example.md is a worked example grounding the expected output; evals/cases.yaml defines evaluation cases; agents/openai.yaml configures the agent's model and parameters.

There is no code to run in the traditional sense. The "runtime" is the Codex agent that reads these files and performs the generation.

How It Is Wired

There is no executable entry point. Execution starts when Codex loads the skill and reads SKILL.md — that file is the sole entry point and carries the widest blast radius, since every invocation routes through its instructions and changing it changes all downstream output. There is no module graph, no call chain, and no cycle; the repository has no code paths to trace.

The flow is declarative: SKILL.md instructs the agent → the agent consults references/transistor-example.md as a format and accuracy template → agents/openai.yaml pins the model and generation settings → evals/cases.yaml provides scored test cases for verifying output. The only external effect is the agent's call to the LLM API, which happens outside this repo. The wiring for anything beyond that — how a user triggers the skill, how evals are run, how output is persisted — has not been mapped in this repository.

File responsibilities: SKILL.md owns the instructions; references/transistor-example.md owns the output template; agents/openai.yaml owns model configuration; evals/cases.yaml owns verification criteria.

How To Use It

There is no README, so no documented commands exist. Based on the structure, install the directory into your Codex skills path and invoke the skill by name in a Codex session. No build step, dependencies, or environment variables are present.

# No documented install command exists — copy the directory into your Codex skills path
# and invoke the skill by name. The README that would confirm this is absent.

Missing evidence: no setup instructions, no run command, and no configuration guide. The agents/openai.yaml file likely holds API-related settings, but its schema is undocumented.

Real-World Use

A technical writer preparing a lesson on semiconductor physics invokes this skill in Codex with a prompt like "create an infographic explaining how a bipolar junction transistor switches." The agent reads references/transistor-example.md, matches that structure and accuracy bar, and produces an SVG diagram with correct terminal labels, current directions, and bias conditions — then evals/cases.yaml checks the result against known-good criteria.

Code Health & Issues

Deep static analysis has not run for this repo. Structure-based observations:

  • Med — no test files; evals/cases.yaml is the only verification artifact and its execution path is undocumented — repository-wide
  • Med — no CI/CD pipeline; no automated gate exists — .github/ absent
  • Med — no LICENSE; usage and redistribution rights are unclear — root
  • Low — no README; intent and onboarding are undocumented — root

The Bottom Line

A small, focused skill definition that solves a real problem: keeping LLM-generated infographics technically honest. It is useful to anyone running Codex who needs consistent educational diagrams, but the missing README, LICENSE, and documented eval workflow limit adoption and make the skill's behavior hard to verify. Forked from cneuralnetwork/generate-technical-infographic; expect to read SKILL.md carefully to understand what it actually promises.