skill

share-explanation

Answers "how does X work" by reading the actual source first, then synthesizing — never from memory.

Locate templates and load SOCIAL.md platform/tone defaults
Parse the target name and any --flags from the request
Resolve source via a 5-tier ladder: skill, command, file, symbol, keyword
Synthesize a structured explanation from only the matched files
Scrub for secrets — a BLOCKED result is a hard stop
Draft platform-aware copy that leads with insight, not process
Reuse-check docs/media/social before regenerating
Populate feature-card (skill) or quote-card (concept), then save
Screenshot via Playwright, then deliver explanation, copy, and PNG

The hardest part of "explain how X works" isn't the writing. It's making sure the explanation matches the code instead of your memory of it.

The share-explanation skill in our social-media-tools plugin solves this by reading the source first, then writing. Ask it "how does X work" and it runs a five-tier resolution ladder to find the real files before saying a word:

1. Exact skill match (SKILL.md)
2. Command file
3. Any source file by name
4. Function or symbol grep
5. Keyword / concept grep as a last resort

It stops at the first hit, so cheap exact matches win before expensive searches — and the search is bounded to the git root, never wandering outside the project.

Then it hands the synthesized explanation to the same pipeline every share-* skill uses: a security scrub that hard-stops on any leaked secret, platform-aware copy, a dark-mode card, and a saved file you can reopen later.

One design choice stands out: the explanation is generated from the files, not from the model's assumptions. That's the difference between docs that drift and docs that stay true.

Run it with:
/social-media-tools:share-explanation <target>

Follow for more Claude Code plugin internals like this.

#ClaudeCode #DeveloperTools #AITools #Plugins

Twitter/X

Explaining your own tooling drifts the moment you write from memory.

The share-explanation skill reads the source first — a 5-tier file-resolution ladder finds the real files, then synthesizes. Code-true docs, not vibes.

#ClaudeCode #DeveloperTools

Bluesky

The share-explanation skill (part of the agentics social-media-tools plugin) won't explain a component from memory — it reads the source first. A five-tier resolution ladder finds the actual files, then it synthesizes. Explanations that match the code, not your assumptions.

Substack

Most docs carry a quiet bug: they're written from memory, not the code, so they drift the day after you write them. The share-explanation skill takes the opposite path — it resolves the real source files through a five-tier ladder (exact skill match down to a keyword grep), reads them, then synthesizes. It reuses the same scrub-copy-card-save pipeline as every share skill, so explaining a component and sharing it are one motion. Subscribe for more on Claude Code plugins that document themselves.