[sreejithraman]

explain is now the one general standard for explaining things, and eli5 is "Explain, as an HTML page with pictures", at the same depth.

What changed

• explain: it explains the requested topic, the last answer, or the item the user points to, for a capable person who lacks this context.
  • It keeps the technical substance, uses everyday words, and names each real term once.
  • It says what the thing is and how it actually works. When they apply, it also covers how it runs or gets triggered, where it lives, and what happens when it fails.
  • It connects to what the user already knows, uses one concrete example from their case for abstract ideas, and gives the tradeoff when there's a choice.
  • It stays short, separates what it knows from what it infers, and says so rather than inventing a link.
• eli5: "Do Explain as an HTML page that leads with big pictures and few words. Make the pictures teach the mechanism. Keep Explain's depth." It shows the page through show-me instead of naming an HTML artifact. Upstream's "someone who knows nothing" is replaced, which addresses "eli5 is sometimes too simple".
• Unchanged: invocation. explain stays user-invoked only in Claude Code and Codex, and eli5 keeps its triggers.
• SOURCES.md: records the current differences for both skills.

Why

A review of past sessions found 13 explicit uses and 11 plain asks. Nearly all explained the conversation's own work. Follow-ups kept asking for the same missing pieces:

• how it runs or where it lives (3);
• the failure path (1);
• connecting context (1);
• a concrete example (2);
• one invented link the user had to correct.

Both skills were used for the same kinds of things. The user wants to pick the format.

Verification

• python3 scripts/check.py passes.
• Codex dry runs on a frozen copy, with outcomes written first. All matched:
  • /explain on a just-shipped change: about 190 words, context first, mechanism, failure path, one example, tradeoff.
  • /explain that on an abstract line: a concrete example, with an ambiguous phrase flagged instead of guessed.
  • /eli5: a phone-width page outline at the same depth.
  • A plain "explain how X works" routes to how.
• Claude Code firing test (claude plugin eval, Opus 5.5, 2 runs per case, installed skills isolated, $6.80):
  • /eli5 and "picture explainer" fired eli5 and read explain in 4 of 4 runs.
  • The near miss fired how in 2 of 2.
  • No page was rendered, because the sandbox denies writes.
• Review: Primary (GPT-6.1 Sol, high) found no implementation issues. It asked for host-listing proof, which the firing test supplied. Gemini 3.8 Flash passed. The OpenCode slot was skipped because no model was usable.

🤖 Generated with Claude Code