Proposal: one namespace — consolidate benchmark and audit into qe, flatten the skill list

[mmcky]

This absorbs the "naming convention, once there is something to generalise from" entry from FUTURE-IDEAS.md (retired 2026-08-13) — the observation it was waiting for has now been made, from the maintainer side: three plugin prefixes is two too many for a catalog this size, and the prefix split (qe: vs benchmark: vs audit:) encodes an installation distinction users don't care about when they're typing a command.

Goal

A single, small, consolidated list of well-named skills under one namespace, so every invocation visibly reads as a QuantEcon skill: /qe:copilot-review, /qe:benchmark, /qe:audit-issues. The catalog is expected to stay small — and to shrink over time as models absorb what today needs a skill — so the organising unit should be the skill name, not a category taxonomy.

Constraint worth knowing first

Claude Code always prefixes a plugin's skills with the plugin name, so truly bare /copilot-review is not available to a marketplace plugin — the shortest stable form is /<plugin>:<skill>. That makes the design question: one plugin named qe, well-named skills inside it. (On older Claude Code versions skills also register bare, but that's a compatibility behaviour, not something to design for.)

Proposed end state

One plugin, qe, in the quantecon marketplace:

Today	Proposed	Notes
/qe:check-style (+ 6 per-category sub-skills)	/qe:check-style	Keep the umbrella; drop the six check-<category> sub-skills. The umbrella already takes categories as arguments (/qe:check-style lecture.md figures math), and natural-language invocation ("check this lecture's figures") still routes there. Six menu entries for one procedure is most of the current menu bloat.
/qe:copilot-review	/qe:copilot-review	Unchanged — already the right shape.
/benchmark:review-acceleration	/qe:benchmark	Matches how everyone already refers to it ("the benchmark skill"); triage vs review is mode selection the skill already does from its arguments. Alternative if the verb matters: /qe:review-acceleration.
/audit:issues	/qe:audit-issues	The audit- stem keeps the family greppable if /qe:audit-prs, /qe:audit-translations etc. ever pass the validation gate (#16) — without needing a plugin to hold them.

Result: four visible commands instead of ten.

What consolidation costs, honestly

• The plugin is the enable unit, and today that's used deliberately: audit is excluded from lecture-repo auto-install so authors don't see org-audit tooling in their menu (README § Installation). One plugin means every consumer gets every skill. I think that's now the right trade — the cost is two extra read-only entries in a four-item menu, not a hazard — but it is the one real regression, and this issue is the place to disagree.
• One version stream. Every change to any skill bumps qe. That's simpler than three streams (and the version-bump CI guard gets simpler too); the changelog just gains section discipline. New version should start strictly above every retiring stream (e.g. 0.5.0 > qe 0.2.2, benchmark 0.3.x/0.4.0, audit 0.2.0) so no number in the merged changelog ever names two trees.
• Installed users must uninstall the retired plugins (claude plugin uninstall benchmark@quantecon audit@quantecon) or stale skills linger under the old names; lecture repos' .claude/settings.json drop benchmark@quantecon from enabledPlugins. Both are one-time and documentable in the changelog entry.

Mechanics (one PR, mostly mechanical)

• Move benchmark/skills/review-acceleration → qe/skills/benchmark, audit/skills/issues → qe/skills/audit-issues; rename skill frontmatter name: to match directories.
• Move benchmark/scripts/ and audit/{references,scripts}/ under qe/ in namespaced subdirectories; update ${CLAUDE_PLUGIN_ROOT} paths in the moved SKILL.md files — this is the step to test from a real consuming project, since path bugs only surface from an install location.
• Fold benchmark/CHANGELOG.md and audit/CHANGELOG.md into qe/CHANGELOG.md as clearly-marked historical sections (their old version numbers and tags — audit--v0.2.0 etc. — remain valid archaeology).
• marketplace.json: remove the two plugin entries; bump the catalogue's top-level version (it moves when a plugin is added or removed).
• Update myst.yml (the docs site renders benchmark/README.md and audit/references/*.md in place), README, CATALOG, using-skills, developing-skills.

Sequencing

After #33 merges. It's open with CI green awaiting review, and a tree-level move underneath it would force a painful rebase of a PR that's already in someone's queue. The consolidation is then a single squash-merged PR.

[mmcky]

Status update, 2026-08-25: this is now unblocked and next in the queue. The sequencing gate ("after #33 merges") cleared on 2026-08-13, and the direction was re-confirmed by @mmcky today — the qe: prefix reading as "this is a QuantEcon skill" is exactly the property to keep, with the audit and benchmark suites folded in flat under it.

One simplification since this was written: PR #52 (in review) removes all seven check-* skills as scaffolding, per the rule that unbuilt work lives in its tracking issue (#3). That makes the first row of the proposal table moot — there is nothing style-shaped to keep or drop, and when the style skills re-land operational they go straight into the consolidated one-umbrella shape proposed here. The remaining mechanics are unchanged: move benchmark and audit under qe/, update ${CLAUDE_PLUGIN_ROOT} paths (tested from a real install), fold the changelogs as historical sections, prune marketplace.json, and sweep myst.yml/README/CATALOG/using-skills/developing-skills.

Plan: land #52 first, then this as a single squash-merged PR.