Action Plans
Action Plans provide a structured remediation prompt for every scanned page. They can be handed directly to an AI coding agent (e.g. Cursor, Windsurf, GitHub Copilot) to implement performance, accessibility, and security fixes.
How it works
Section titled “How it works”PerfLeaf generates action plans through a five-stage deterministic pipeline:
- Normalisation — Raw page data (Core Web Vitals, assets, accessibility, security, DOM, GEO, SEO) is normalised into a unified issue list with consistent categories.
- Scoring & Ranking — Each issue receives an impact score (0–100) based on metric deltas, category effort, grade boosts, and regression boosts. Issues are sorted by composite score (impact ÷ effort).
- Recommendation Enrichment — Deterministic recommendations are merged into scored issues, adding remediation actions and learn-more URLs.
- Prompt Generation — A structured
AIAgentPromptis built with full page context and prioritised tasks. - AI Enrichment (paid plans only) — Tasks are sent to the AI service for improved descriptions, code examples, and additional remediation steps.
Creating an action plan
Section titled “Creating an action plan”From a Page Report:
- Click Generate Action Plan.
- PerfLeaf runs the full pipeline and stores the result.
- You’re taken to the action plan view with the prioritised task list.
Task structure
Section titled “Task structure”Each task in the action plan includes:
| Field | Description |
|---|---|
| Issue Category | e.g. render_blocking, javascript_optimisation, accessibility_critical |
| Priority Rank | Sorted by composite score |
| Measurable Goal | Specific metric improvement target |
| Technical Objective | What to implement |
| Performance Target | Quantified success criteria |
| Constraints | What must not break |
| Acceptance Criteria | Verification steps |
| DOM Selectors | Relevant elements (when available) |
| Regression Risks | Side-effects to watch for |
| Remediation Actions | Step-by-step fix instructions |
| AI Hints | Code examples from deterministic + AI enrichment |
| Learn More URL | Reference documentation |
Scoring transparency
Section titled “Scoring transparency”Every action plan includes a scoring breakdown showing:
- Model version and weight configuration
- Per-issue explanation strings (e.g.
base=45, sizeImpact=+12, gradeBoost=+8)
This ensures full transparency into how issues are prioritised.
Scoring boosts
Section titled “Scoring boosts”| Boost | Condition | Effect |
|---|---|---|
| Grade boost | Category has a C or D grade | +15% to impact score |
| Regression boost | Metric degraded vs. previous report | +10% to impact score |
AI enrichment (plan-gated)
Section titled “AI enrichment (plan-gated)”For paid plans with the aiActionPlanEnrichment feature enabled, the top tasks are sent to the AI service which returns:
- Improved descriptions tailored to the detected tech stack
- Code examples specific to the framework in use
- Additional remediation steps
If the AI service is unavailable, the deterministic plan is stored unchanged.
| Feature | Sprout (Free) | Grow | Bloom |
|---|---|---|---|
| Deterministic action plan | ✓ | ✓ | ✓ |
| Recommendation enrichment | ✓ | ✓ | ✓ |
| AI enrichment | ✗ | ✓ | ✓ |
Navigating action plans
Section titled “Navigating action plans”Action plans are accessible at /dashboard/action-plans/[id]. From there you can:
- View the full prioritised task list
- Navigate to the AI Coding Agent Prompt
- Return to the source page report
AI Coding Agent Prompt
Section titled “AI Coding Agent Prompt”Each action plan has an associated Prompt Generation page that formats the plan as a prompt ready for AI coding agents. See Prompt Generation for details.