Fix Packs and the AI Fix Workflow
From finding to AI-ready fix to verified result. Protected changes always need a human.
A Fix Pack turns a finding into a structured, AI-ready change. The full Fix Pack for one URL is a one-time £39 product bound to that report: one URL, one fix-pack generation, one verification window of 30 days.
The two-sided AI Fix object
HubSEO's signature "Create AI fix" action morphs in place into a dual-sided object:
Your side. The problem and the goal in plain English: what is happening, why it matters, and what success looks like.
The AI agent side. A deterministic, URL-scoped specification:
- Target area: the file, template, or CMS surface to change.
- Technical outcome: the measurable end state.
- Constraints: what must not be touched.
- Steps: the ordered change sequence.
- Tests: how the agent proves the change worked.
The object is scoped to the audited URL and backed by measured evidence: Lighthouse lab selectors, HTTP status codes, or verified GSC data. Prompts never contain placeholder domains or generic templates.
Supported agents
The fix is written to be consumed by Cursor, Claude Code, Copilot, Codex, or Aider. Any competent coding agent that can read a structured brief and edit your repo applies them equally well.
Three tiers of intervention
Who executes: you review the prepared change. Verification: re-scan when you apply it yourself. The default surface for Fix Pack purchases: HubSEO prepares everything, nothing executes until you act.
Safe autonomous mode
Autonomy is opt-in and fenced:
- Default state is strictly off:
auto_deploy_safe_changes: false. - Even when enabled, deployments are capped at 3 per week.
- Only changes classified as
safe_autowith confidence ≥ 0.85 and a verified rollback plan can execute without a manual click. - Protected directives can never execute autonomously: canonical tags, 301
redirects,
robots.txt, URL migrations. These always require explicit human verification. This protects your uptime, deliberately.
See Autonomous mode for the operating procedure.
Closed-loop verification
"Don't trust the fix. Verify it." When a fix is applied, HubSEO re-scans and compares evidence on the finding's own terms. A finding closes only when the re-scan confirms the measurement changed. The verification badge reads "Evidence-backed · Live re-scan".
What verification does not claim: a merged pull request proves nothing by itself. Only the re-scan does.