/optimize — standards checks that become checkbox-picked plans

Every project accumulates the same small gaps: no robots.txt, no llms.txt, a single stray favicon.ico instead of a real icon set, no dependency-update bot. /optimize scans for the standards that actually apply to this project's type, and turns the gaps into .plans you pick from — not a wall of text you have to triage by hand.

The gap

Unlike security holes (/audit's job), hygiene and discoverability gaps don't page anyone — they just quietly cost a project its social preview image, its crawler visibility, or its "why doesn't this repo have a SECURITY.md" moment. What was missing was a session that:

1. Detects project type before suggesting anything — a CLI tool never gets told to add Open Graph tags; a static site never gets told to add CODEOWNERS alone as if that were the priority
2. Checks a reproducible baseline instead of relying on model recall alone
3. Ranks by impact/effort, caps the noise, and lets a human pick a subset instead of writing ten plans nobody asked for
4. Writes plans only — it never touches robots.txt or wires an OG image pipeline itself

What /optimize does

One project, one scan. Pipeline (fixed):

resolve project → detect project type(s)
  → build applicable-standards checklist → check presence/absence
  → rank + cap at 10 candidates → present findings (human)
  → checkbox confirm (or --write / --dry-run)
  → emit chosen candidates → stop

Signal	Type	Checks
SSR framework / bundled index.html	Web app	OG/Twitter cards, robots.txt/sitemap.xml/llms.txt, PWA icons, structured data
package.json with bin, published library	CLI / library	CODEOWNERS, SECURITY.md, CONTRIBUTING.md, release config, .editorconfig
Docusaurus/MkDocs/Sphinx/VitePress markers	Docs	llms.txt/llms-full.txt, search config
Any repo	General	dependency bot, LICENSE, CHANGELOG, issue/PR templates

A project can match more than one row — Anchor's own docs site, for instance, reads as Docs + CLI/library + General all at once, and only the union of those categories gets checked.

Flag	Role
--dry-run	Findings only; zero plan files
--write	Skip confirm; write all presented candidates
--to features / --to bugs	Opt-in straight to a ready lane
--continue	Resume a capped backlog from a prior run

The checklist is a floor, not a ceiling

A hardcoded baseline keeps results reproducible run to run — but "popular standard" keeps shifting past any model's training cutoff. /optimize may propose items beyond the baseline when it has good reason to, always tagged (emerging — verify still current) so the human knows it wasn't checklist-verified.

No refuse gate

/optimize is hygiene and DX, not security — there's no frontier/reasoner wall the way /audit has one. It states a soft mid, reasoner preference (judgment about "is this standard actually relevant here" benefits from a stronger model) and runs on any tier that picks it up.

Where it ships

• Claude: .claude/commands/optimize.md
• Grok: .grok/skills/optimize/SKILL.md
• Docs: Skills → /optimize
• Scaffolded with the other dual-use skills (scripts/anchor.py) — it installs into every project scaffolded or reconciled with Anchor, not just this repo

Written plans default to sticky .plans/drafts/opt-<kebab>.local.md — review and promote with /draft --promote, or execute directly with /work once they land in a ready lane.