OPEN SOURCE · 13 SKILLS · MIT

A fix isn't finished until the next audit says so.

Groundcrew is 13 open-source Markdown skills that guide your coding agent through auditing a site, fixing one verified defect in your repo, and checking it again after deploy.

locally verified — the tests pass

deployed — the change is live

audit-verified — a fresh measurement confirms closure

The rules live in the repository, so you can inspect them before your agent runs them.

INSTALL · PINNED TO v0.2.0
REAL OUTPUT
curl -fsSL https://raw.githubusercontent.com/techwright-lab/groundcrew/v0.2.0/install.sh | bash
Groundcrew install into ~/.claude/skills
  ai-visibility · authority-review · backlink-opportunities
  competitor-watch · content-desk · content-strategy
  eeat-review · fix-my-site · keyword-scout · score-report
  site-audit · standup · trustgrowth
Groundcrew doctor
  ✓ evidence schema parses
  ✓ checked 13 skill frontmatters and shared references
passed=1 errors=0
Next: run a local workflow now, for example:
  "inspect this repo and fix one verifiable site defect."
TrustGrowth is optional; connect it later for persisted evidence,
prioritization, scheduled work, and re-audit verification.

what you review is what you run

browse all 13 skills →

MIT licensed. No account or API key required. Connect TrustGrowth later for live findings, scheduling, history, and re-audits.

v0.2.0 · github.com/techwright-lab/groundcrew

Audits find problems. Groundcrew closes one.

Most audit findings become tickets. An audit can tell you thirteen pages are missing author markup — it cannot open your layout. The thing that can open your layout is already sitting in your terminal, and it reads Markdown.

One finding. One pull request. One fresh audit.

REBUILT

STATION 1 · FOUND

Choose one code-fixable finding and verify it against the live site and current repository.

STATION 2 · FIXED

Change the root cause, run the project's tests and build, and prepare a pull request.

STATION 3 · VERIFIED

Record local verification, deployment, and audit verification separately. Never call the finding resolved early.

↻ RE-AUDIT · a fresh measurement returns the closed finding to FOUND — or retires it

THE LOOP · It reads evidence from TrustGrowth, an import, or the public site. Changes go to your repository.

The loop closes in your repository, where the fix and its tests live.

Thirteen skills, readable before you run them.

fix-my-site FLAGSHIP
standalone public/repo defect → tested PR connected prioritized queue + re-audit closure
site-audit
standalone point-in-time audit from what's available connected scheduled history, severity
ai-visibility
standalone crawler access by each bot's documented purpose; readiness vs. visibility kept honest connected the measured funnel, when the live API documents it
content-strategy
standalone evidence-grounded pillars and sequence, owner-approved connected measured gap inputs
authority-review
standalone entity and reference standing; refuses to invent a score connected TrustGrowth's own authority pillar, named as such
backlink-opportunities
standalone profile → human-reviewed plan; never disavow, never automated outreach connected detailed reads when documented
keyword-scout
standalone imports or cost-gated research → prioritized plan connected normalized opportunities
competitor-watch
standalone current observations, never fabricated movement connected curated tracking history
eeat-review
standalone observable trust-signal checklist connected persisted assessment
score-report
standalone source-labeled report connected score history + the publication-safe packet
standup
standalone what the evidence says happened connected persisted team activity
content-desk WAITLIST
standalone read-only inventory connected writes stay dark, Growth only
trustgrowth
standalone the optional provider connector — last on the list, deliberately connected the complete managed layer

Thirteen files — read every one before you run it. A skill is a procedure you can audit, not a model you have to trust. One evidence contract behind all of them, checked by a dependency-free doctor script.

Rules you can inspect.

“Slop borrows against domain trust. Evidence compounds it.”

WHY-NOT-SLOP.md · committed in the repo · MIT

These rules ship with the skills. Read them before you install, and verify what the doctor checks.

“Numbers are measurements or they are not numbers.”

null means unknown and is reported as unknown — never zero, never interpolated, never smoothed into a nicer curve.

ETHICS.md · principle 3

A principle nobody can check is decoration; we don't ship those.

WHAT THE SKILLS REFUSE

default branch
never pushed to, unless your workflow explicitly allows it
robots · redirects · canonicals · content judgment
owner confirmation, every time; those can deindex a site
billable research
no call without a bounded cost estimate and explicit approval for that batch
disavow · toxicity verdicts · automated outreach
never
unmappable findings
escalated for human review, not guessed
fetched web content
treated as untrusted evidence, never as agent instructions

Free to run. Better with live evidence.

Groundcrew needs no account or API key. It works from your public site, your repository, and imported evidence.

Connect TrustGrowth when you want measured findings, priority queues, history, scheduled work, and post-deploy re-audits.

STANDALONE

  • All 13 Markdown skills
  • Public site, repository, imports
  • No signup · MIT

CONNECTED

  • Live TrustGrowth findings
  • Prioritization and history
  • Scheduled work
  • Post-deploy re-audits

Connected starts at $9/mo (Starter) — the API key that turns inspection into measurement. Plans and API limits → Pricing.

WHAT THIS DOES NOT SAY

Skills are plain Markdown, and file compatibility is not feature parity: hosts differ, and REST and MCP coverage can differ too — the skills discover the live contract instead of assuming it.

Structured errors, no guesswork.

FROM THE API

Connected mode returns structured errors your agent can act on. Each response names the problem and includes the fields needed to recover.

401 missing_token

403 plan_limit · with upgrade_to

403 insufficient_scope · with required_scope

403 coming_soon · with waitlist_url

402 insufficient_credits · with balance / required / shortfall / topup_url

429 rate_limit_exceeded · with the rate-limit headers

Branch on the error code, not the HTTP status. Full response shapes live in the API reference.

Install Groundcrew

curl -fsSL https://raw.githubusercontent.com/techwright-lab/groundcrew/v0.2.0/install.sh | bash

Get a score to work from

Start with the open skills. Add a TrustGrowth API key later if you want live findings and scheduled verification.