feat: improve skill scores for 5 lowest-scoring skills

[rohan-tessl]

Hey @doodledood 👋

I ran your skills through tessl skill review at work and found some targeted improvements.

Here's the full before/after:

Skill	Before	After	Change
clean-slop	38%	89%	+51%
investigate-bug	36%	74%	+38%
audit-ux	39%	74%	+35%
review-prompt	38%	69%	+31%
update-claude-md	18%	18%	+0%*

*update-claude-md contains the reserved word "claude" in its name, which causes the tessl validator to fail deterministically and skip the LLM judge entirely. The content improvements are genuine (expanded description with "Use when..." clause, structured best-practices sections, verification checklist) but the score can't reflect them. You may want to consider renaming (e.g. update-project-instructions) or accept the validation trade-off — totally your call.

This PR is intentionally scoped to the 5 lowest-scoring skills. More skills can be improved in follow-ups or via automated review on future PRs.

Changes summary

All 5 skills received:

• Expanded frontmatter descriptions with specific concrete actions and explicit "Use when..." trigger clauses
• Removed unknown context: fork frontmatter keys (caused validation warnings on 3 skills)

investigate-bug: Added structured workflow (gather symptoms → identify scope → invoke consultant → relay findings), expected output format with root cause, execution flow, blast radius, and regression test plan sections.

review-prompt: Added 10-Layer Architecture reference table, structured output description (assessment, score, strengths, issues table with AUTO_FIXABLE/NEEDS_USER_INPUT tags), clarified read-only constraint.

clean-slop: Added four concrete slop categories with examples (useless comments, verbose docs, filler phrases, unnecessary patterns), scope rules (file paths vs git diff), decision rule for remove/keep, and output format description.

audit-ux: Added 4-step workflow (gather references → identify changes → systematic review → report findings), five issue categories with specific examples, and structured output format description.

update-claude-md: Restructured best practices into scannable sections (structure, length budget table, progressive disclosure, do/don't lists, examples), added verification checklist.

Honest disclosure — I work at @tesslio where we build tooling around skills like these. Not a pitch - just saw room for improvement and wanted to contribute.

Want to self-improve your skills? Just point your agent (Claude Code, Codex, etc.) at this Tessl guide and ask it to optimize your skill. Ping me - @rohan-tessl - if you hit any snags.

Thanks in advance 🙏

[gemini-code-assist]

Code Review

This pull request refactors and modularizes documentation across several modules, including Electron, Slack, Vercel Sandbox, Docker, and React. It introduces dedicated reference files for detailed tasks and configurations, keeping the primary SKILL.md files concise and focused on core patterns. Key improvements include enhanced error handling in Vercel Sandbox code snippets, more robust environment detection in Docker scripts, and the addition of practical code examples for React performance optimization. Feedback suggests improving command result validation and regex robustness in the Vercel Sandbox documentation.

[gemini-code-assist]

The runCommand result is parsed as JSON without checking the exit code. If the command fails, JSON.parse will throw an unhelpful error. Please verify the exit code before parsing, consistent with the check on line 133.

Suggested change

	const ssResult = await sandbox.runCommand("agent-browser", [
	"screenshot", "--json",
	]);
	const ssPath = JSON.parse(await ssResult.stdout())?.data?.path;
	const b64Result = await sandbox.runCommand("base64", ["-w", "0", ssPath]);
	const ssData = JSON.parse(await ssResult.stdout());
	const ssResult = await sandbox.runCommand("agent-browser", [
	"screenshot", "--json",
	]);
	if ((await ssResult.exitCode()) !== 0) {
	throw new Error(`Screenshot failed: ${await ssResult.stderr()}`);
	}
	const ssData = JSON.parse(await ssResult.stdout());

[gemini-code-assist]

The regex \w+ for the element role is restrictive and may fail on kebab-case roles (e.g., tree-item). Using \S+ is more robust for matching any non-whitespace role name.

Suggested change

	const match = line.match(/^(@e\d+)\s+\w+\s+"(.+)"/);
	const match = line.match(/^(@e\d+)\s+\S+\s+"(.+)"/);