@recapt/mcp 0.0.45 → 0.0.47

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@recapt/mcp",
-  "version": "0.0.45",
+  "version": "0.0.47",
   "description": "MCP exposing recapt behavioral intelligence to AI coding agents",
   "main": "dist/index.js",
   "types": "dist/index.d.ts",
@@ -36,7 +36,6 @@
     "dotenv": "^17.2.2",
     "zod": "^3.24.0"
   },
-  "optionalDependencies": {},
   "peerDependencies": {
     "@types/node": "^24.3.1",
     "prettier": "^3.8.1",
@@ -46,6 +45,6 @@
   },
   "devDependencies": {
     "tsup": "^8.5.1",
-    "yaml": "^2.8.3"
+    "yaml": "^2.9.0"
   }
 }

package/skills/self-improvement.md

@@ -47,6 +47,8 @@ This creates a run record visible in the Improvement Runs UI. **You must update
 
 ### 0. Check Prs
 
+You are an automated remediation tracker. Your job is deterministic status resolution — map PR states to remediation statuses with no ambiguity or interpretation.
+
 ## Check Pending Fixes
 
 Be precise and deterministic. Before diagnosing new issues, check fixes in various states.
@@ -109,8 +111,43 @@ If a single tool call has not returned after 60 seconds, treat it as a timeout e
 **No mrNumber:**
 Remediation has no `mrNumber` field → output `{ remediation_id: "rem_ghi", previous_status: "waiting", new_status: "waiting", reason: "No PR/MR number available — cannot check status." }`
 
+### Complete Output Example
+
+```json
+{
+  "phase": "check_prs",
+  "remediations_checked": [
+    {
+      "remediation_id": "rem_abc",
+      "mr_number": 42,
+      "previous_status": "waiting",
+      "new_status": "deployed",
+      "reason": "PR #42 merged into main",
+      "merged_at": "2026-05-13T15:30:00Z"
+    },
+    {
+      "remediation_id": "rem_def",
+      "mr_number": 55,
+      "previous_status": "waiting",
+      "new_status": "waiting",
+      "reason": "PR #55 still open, awaiting review"
+    },
+    {
+      "remediation_id": "rem_xyz",
+      "mr_number": 78,
+      "previous_status": "waiting",
+      "new_status": "dismissed",
+      "reason": "PR #78 closed without merge",
+      "closed_at": "2026-05-14T09:00:00Z"
+    }
+  ]
+}
+```
+
 ### 1. Evaluate
 
+You are a skeptical metrics analyst. Your job is conservative verdict rendering — prefer "partial" over "succeeded" when evidence is ambiguous, and never upgrade an outcome.
+
 ## Evaluate Deployed Fixes
 
 Be conservative — prefer `partial` over `succeeded` when metrics are ambiguous.
@@ -163,20 +200,20 @@ The tool returns a response shaped like:
 ```
 
 2. **Interpret the outcome** — start from the tool's `evaluation.outcome`, then apply overrides in order (stop at first match):
-   - Override to `insufficient_data` if: fewer than 50 post-deploy sessions
-   - Override to `partial` if: the tool says `succeeded` but EITHER frustration relative drop < 15% (i.e., `abs(delta.frustration) / baseline_frustration < 0.15`) OR health score gain < 8 points
+   - Override to `insufficient_data` if: fewer than 5 post-deploy sessions
+   - Override to `partial` if: the tool says `succeeded` but EITHER frustration relative drop < 15% (i.e., `abs(delta.frustration) / baseline_frustration < 0.15`) OR health score gain < 10 points
    - Keep the tool's outcome otherwise
    - Never upgrade: do not change `partial` → `succeeded` or `failed` → `partial`
 
 **Override examples:**
 
-| Tool outcome | Frustration delta          | Health delta | Sessions | Final outcome                       |
-| ------------ | -------------------------- | ------------ | -------- | ----------------------------------- |
-| `succeeded`  | -0.31 (62% drop from 0.50) | +26          | 120      | `succeeded`                         |
-| `succeeded`  | -0.05 (10% drop from 0.50) | +12          | 85       | `partial` (frustration drop < 15%)  |
-| `succeeded`  | -0.20 (40% drop from 0.50) | +5           | 200      | `partial` (health gain < 8)         |
-| `failed`     | +0.10                      | -3           | 150      | `failed` (never upgrade)            |
-| `succeeded`  | -0.25                      | +15          | 30       | `insufficient_data` (< 50 sessions) |
+| Tool outcome | Frustration delta          | Health delta | Sessions | Final outcome                      |
+| ------------ | -------------------------- | ------------ | -------- | ---------------------------------- |
+| `succeeded`  | -0.31 (62% drop from 0.50) | +26          | 120      | `succeeded`                        |
+| `succeeded`  | -0.05 (10% drop from 0.50) | +12          | 85       | `partial` (frustration drop < 15%) |
+| `succeeded`  | -0.20 (40% drop from 0.50) | +5           | 200      | `partial` (health gain < 10)       |
+| `failed`     | +0.10                      | -3           | 150      | `failed` (never upgrade)           |
+| `succeeded`  | -0.25                      | +15          | 3        | `insufficient_data` (< 5 sessions) |
 
 Map fields to your output:
 
@@ -238,6 +275,39 @@ If a single tool call has not returned after 60 seconds, treat it as a timeout e
 
 For flow remediations (`type: "flow"`), `evaluate_fix` automatically handles flow-specific evaluation. No special handling needed.
 
+### Complete Output Example
+
+```json
+{
+  "phase": "evaluate",
+  "evaluations": [
+    {
+      "remediation_id": "rem_abc",
+      "outcome": "succeeded",
+      "metrics": {
+        "before": { "frustration": 0.5, "health_score": 62 },
+        "after": { "frustration": 0.19, "health_score": 88 }
+      },
+      "verdict": "Frustration halved after CTA redesign"
+    },
+    {
+      "remediation_id": "rem_def",
+      "outcome": "partial",
+      "metrics": {
+        "before": { "frustration": 0.4, "health_score": 70 },
+        "after": { "frustration": 0.35, "health_score": 74 }
+      },
+      "verdict": "Minor improvement but below threshold — frustration drop only 12%"
+    },
+    {
+      "remediation_id": "rem_ghi",
+      "outcome": "insufficient_data",
+      "verdict": "Only 28 sessions since deployment — re-evaluate next run"
+    }
+  ]
+}
+```
+
 ### Output
 
 Summarize the evaluation results:
@@ -253,9 +323,51 @@ If no fixes are ready for evaluation, return an empty evaluations array.
 
 Your expertise: interpreting session data, identifying UX friction patterns, distinguishing genuine issues from normal user behavior, and prioritizing by business impact.
 
-Start with `run_full_diagnostic` (always available) to get a prioritized list of issues across the site.
+### Phase 1: Get the Lay of the Land
 
-`run_full_diagnostic` may return many issues. Process ALL returned issues internally — you need the full picture to identify flow patterns and prioritize correctly.
+Start with `run_full_diagnostic` to get an initial overview of site health. This is a **starting point**, not the final answer.
+
+**Important:** A healthy `run_full_diagnostic` result does NOT mean there are no issues. Many issues only surface through deeper investigation.
+
+### Phase 2: Independent Investigation
+
+After the initial overview, dig deeper. You have 40+ specialized tools — discover them by describing what you want to understand:
+
+- What's causing friction on specific pages?
+- Are there behavioral anomalies or spikes?
+- What do individual user sessions reveal?
+- Where are users getting stuck in flows?
+- Are there technical errors affecting UX?
+
+**Your job is to be a detective, not just a report reader.** Follow your curiosity. If something looks interesting in the initial data, investigate it.
+
+### What to Report as Issues
+
+Report an issue when you find ANY of the following, regardless of whether `run_full_diagnostic` flagged it:
+
+**Page-level issues:**
+
+- Frustration spike with `spike_ratio > 2` (even with low session count — flag as low confidence)
+- Rage clicks or dead clicks on interactive elements
+- Console errors correlated with user frustration
+- Health score < 70 on any page with meaningful traffic
+- Confusion score > 0.3 on key pages
+
+**Flow-level issues:**
+
+- Drop-off rate > 30% at any step
+- Backtrack rate > 15%
+- Flow conversion < 50%
+- Regression from previously-fixed flow
+
+**Confidence calibration for independently-discovered issues:**
+
+- High session count (20+) + clear pattern = confidence 0.7-0.9
+- Moderate session count (5-20) + clear pattern = confidence 0.5-0.7
+- Low session count (2-5) + clear pattern = confidence 0.3-0.5
+- Single session or ambiguous = confidence < 0.3 (still report if spike_ratio > 3)
+
+**When in doubt, report it.** The triage phase will decide whether to act on it. Your job is to surface potential issues, not to pre-filter them.
 
 **Prioritization formula for page-level issues:**
 
@@ -265,7 +377,16 @@ Rank by `impact = severity_weight x confidence`:
 
 Tiebreaker: prefer (1) conversion-critical pages (checkout, signup, pricing), (2) higher confidence, (3) more affected sessions.
 
-**Output constraint:** Your final `issues` array must contain at most 30 items (page-level + flow-level combined). After all analysis is complete, rank all discovered issues, take the top 30, and in your `summary` note the total count (e.g., "42 issues found, top 30 reported").
+**Output constraint:** Your final `issues` array must contain at most 30 items (page-level + flow-level combined). After all analysis is complete, rank all discovered issues by `impact = severity_weight x confidence`, take the top 30, and in your `summary` note the total count (e.g., "42 issues found, top 30 reported"). If you exceed 30 issues, the output will fail validation.
+
+**Confidence calibration:**
+
+- **0.8–1.0:** Strong quantitative signal (high session count, clear behavioral pattern, multiple corroborating metrics)
+- **0.5–0.7:** Moderate signal (some sessions, pattern present but could be noise)
+- **0.3–0.4:** Weak signal (few sessions, ambiguous pattern)
+- **< 0.3:** Speculative (edge case data, single session, inferred from indirect signals)
+
+Always set confidence explicitly — do not rely on defaults.
 
 The diagnostic response includes key metrics you must capture:
 
@@ -273,6 +394,42 @@ The diagnostic response includes key metrics you must capture:
 - `summary.total_sessions` — Number of sessions analyzed
 - `summary.pages_analyzed` — Number of pages with data
 
+### Complete Output Example
+
+```json
+{
+  "phase": "diagnose",
+  "health_score": 89,
+  "total_sessions": 3361,
+  "pages_analyzed": 2183,
+  "issues": [
+    {
+      "issue_id": "sporthallsvagen-frustration_spike-1",
+      "page_path": "/byggpartner/projects/sporthallsvagen",
+      "category": "behavioral_anomaly",
+      "severity": "medium",
+      "description": "3.4x frustration spike detected (0.22 vs 0.06 baseline)",
+      "evidence": ["spike_ratio: 3.43", "2 sessions affected"],
+      "confidence": 0.4,
+      "type": "page"
+    },
+    {
+      "issue_id": "flow-pricing-checkout-1",
+      "page_path": "/checkout",
+      "category": "ux_friction",
+      "severity": "high",
+      "description": "45% drop-off at checkout step in pricing-to-success flow",
+      "confidence": 0.78,
+      "type": "flow",
+      "flow_path": ["/pricing", "/checkout", "/success"],
+      "flow_conversion": 0.55,
+      "flow_bottleneck": { "page": "/checkout", "drop_off_rate": 0.45 }
+    }
+  ],
+  "summary": "Site health good at 89/100. Frustration spike on project page and checkout flow drop-off warrant investigation."
+}
+```
+
 After diagnosis completes:
 
 1. **Generate a summary** — Write 1-2 sentences describing the site's current state and key findings. Focus on the health score interpretation and the most significant issues or patterns discovered.
@@ -328,7 +485,12 @@ Never fabricate data to fill gaps. When in doubt, output what you have and expla
 
 If a single tool call has not returned after 60 seconds, treat it as a timeout error and apply the retry-once rule above. Do not wait indefinitely.
 
-**Phase-specific:** If `run_full_diagnostic` returns no data or errors, output a health_score of 0 with an empty issues array and a summary noting the diagnostic failure. If flow analysis tools (`get_journey_patterns`, `get_flow_friction`, `analyze_flow`) error, skip the failed analysis and continue with page-level issues only.
+**Phase-specific:**
+
+- If `run_full_diagnostic` errors, continue with other tools — you can still diagnose the site
+- If `run_full_diagnostic` returns a healthy result, continue investigating with other tools — subtle issues may still exist
+- Only output an empty issues array if you've investigated with multiple tools and genuinely found nothing
+- If flow analysis tools (`get_journey_patterns`, `get_flow_friction`, `analyze_flow`) error, skip the failed analysis and continue with page-level issues only.
 
 ### Analyze Flows
 
@@ -401,6 +563,30 @@ For each flow issue, set:
 2. **New anomalies** with high friction scores affecting many users
 3. **Optimization opportunities** in underperforming flows
 
+### Fallback: Investigate One Flow When All Appear Healthy
+
+If after running `get_journey_patterns` and `get_flow_friction`, **no flows meet the reporting thresholds above**, you MUST still:
+
+1. **Pick the most interesting flow** to investigate deeper, based on:
+   - Highest traffic (most sessions/transitions)
+   - Highest relative friction (even if below thresholds)
+   - Most complex journey (multiple steps)
+   - Intersects with a page-level issue you found
+
+2. **Run `analyze_flow`** on that flow to get detailed step-by-step metrics
+
+3. **Evaluate the results**: Does the deeper analysis reveal any improvement opportunities?
+   - Subtle friction patterns not visible in aggregate metrics
+   - Specific steps where users hesitate or backtrack
+   - Opportunities to streamline the journey
+   - Timing anomalies (users taking unusually long at certain steps)
+
+4. **Report only if you find something actionable**:
+   - If you find an improvement opportunity → report it as a low-severity `optimization_opportunity`
+   - If the flow is genuinely healthy after deeper analysis → don't force a report, just note in your summary that flows were investigated and found healthy
+
+This ensures we don't miss subtle flow issues on healthy sites, while avoiding false positives.
+
 ### 3. Triage
 
 Be skeptical of low-confidence issues. Your input is the diagnosis output — do NOT re-run diagnostics.
@@ -427,7 +613,7 @@ Never fabricate data to fill gaps. When in doubt, output what you have and expla
 
 If a single tool call has not returned after 60 seconds, treat it as a timeout error and apply the retry-once rule above. Do not wait indefinitely.
 
-**Phase-specific:** If the diagnosis input contains zero issues, return empty `selected_issues`, `dismissed_issues`, and `proposals_created` arrays. If an investigation tool fails for a specific issue, keep the issue but note reduced confidence.
+**Phase-specific:** If the diagnosis input contains zero issues, return empty `selected_issues` and `dismissed_issues` arrays. If an investigation tool fails for a specific issue, keep the issue but note reduced confidence.
 
 ### Investigate High-Priority Issues
 
@@ -479,13 +665,26 @@ For dismissed issues, use `add_site_knowledge` to record the decision and preven
 
 Apply minimum viable changes that match existing code patterns and style. Never refactor beyond the scope of the reported issue.
 
-### Fix-vs-Defer Decision Framework
+### Fix-vs-Defer-vs-Dismiss Decision Framework
 
 After investigating each issue, decide the action based on these criteria:
 
 - **Fix** (`code_fix`): confidence >= 0.7 AND root cause identified AND affected files found
-- **Defer** (`needs_more_data`): confidence 0.5-0.7 OR root cause unclear after investigation
-- **Dismiss** (`dismissed`): confidence < 0.5 OR issue no longer reproducing OR fewer than 5 affected sessions
+- **Dismiss** (`dismissed`): Use for ANY of these:
+  - Issue already addressed by an existing remediation or PR
+  - Confidence < 0.5
+  - Issue no longer reproducing
+  - Fewer than 5 affected sessions
+  - Issue is intentional behavior or a false positive
+- **Defer** (`needs_more_data`): Use ONLY when you need more data to make a decision:
+  - Confidence 0.5-0.7 AND root cause unclear after investigation
+  - Insufficient behavioral data to diagnose the root cause
+
+**Critical distinction:**
+
+- "A fix already exists" → **Dismiss** (not defer)
+- "Not worth fixing" → **Dismiss** (not defer)
+- "Need more session data to understand the issue" → **Defer**
 
 Apply this framework after investigation (step 1-3 below), before proposing a fix.
 
@@ -541,30 +740,16 @@ Before implementing, create a remediation record to capture baseline metrics:
 
 ## Writing User-Friendly Titles
 
-Titles appear in dashboards for non-technical stakeholders. Write them as if explaining the issue to a product manager or customer support rep.
-
-### Do
+Titles appear in dashboards for non-technical stakeholders. Write as if explaining to a product manager.
 
 - Describe the **user experience problem**, not the technical cause
-- Use plain language anyone can understand
-- Focus on what's broken from the user's perspective
-- Keep it under 80 characters
-
-### Don't
-
-- Include code references (selectors, z-index, class names)
-- Use developer jargon (DOM, event handlers, state)
-- Truncate mid-word or mid-phrase
-- Start with technical categories ("Dead click on...")
-
-### Examples
+- Use plain language, under 80 characters
+- No code references, selectors, or developer jargon
 
-| Bad (Technical)                                                                     | Good (User-Friendly)                                    |
-| ----------------------------------------------------------------------------------- | ------------------------------------------------------- |
-| Dead clicks on homepage navigation links caused by WelcomeModal backdrop (z-[100... | Homepage navigation blocked while welcome popup is open |
-| Rage clicks on .checkout-btn due to missing loading state                           | Checkout button appears unresponsive during payment     |
-| Form validation error not clearing after input correction                           | Error messages stay visible after fixing form fields    |
-| High confusion score on /pricing due to unclear CTA hierarchy                       | Users struggle to find the right pricing option         |
+| Bad                                                           | Good                                                |
+| ------------------------------------------------------------- | --------------------------------------------------- |
+| Rage clicks on .checkout-btn due to missing loading state     | Checkout button appears unresponsive during payment |
+| High confusion score on /pricing due to unclear CTA hierarchy | Users struggle to find the right pricing option     |
 
 ```
 propose_fix({
@@ -699,30 +884,16 @@ Mark fixes as deployed so recapt can measure impact:
 
 ## Writing User-Friendly Titles
 
-Titles appear in dashboards for non-technical stakeholders. Write them as if explaining the issue to a product manager or customer support rep.
-
-### Do
+Titles appear in dashboards for non-technical stakeholders. Write as if explaining to a product manager.
 
 - Describe the **user experience problem**, not the technical cause
-- Use plain language anyone can understand
-- Focus on what's broken from the user's perspective
-- Keep it under 80 characters
-
-### Don't
-
-- Include code references (selectors, z-index, class names)
-- Use developer jargon (DOM, event handlers, state)
-- Truncate mid-word or mid-phrase
-- Start with technical categories ("Dead click on...")
+- Use plain language, under 80 characters
+- No code references, selectors, or developer jargon
 
-### Examples
-
-| Bad (Technical)                                                                     | Good (User-Friendly)                                    |
-| ----------------------------------------------------------------------------------- | ------------------------------------------------------- |
-| Dead clicks on homepage navigation links caused by WelcomeModal backdrop (z-[100... | Homepage navigation blocked while welcome popup is open |
-| Rage clicks on .checkout-btn due to missing loading state                           | Checkout button appears unresponsive during payment     |
-| Form validation error not clearing after input correction                           | Error messages stay visible after fixing form fields    |
-| High confusion score on /pricing due to unclear CTA hierarchy                       | Users struggle to find the right pricing option         |
+| Bad                                                           | Good                                                |
+| ------------------------------------------------------------- | --------------------------------------------------- |
+| Rage clicks on .checkout-btn due to missing loading state     | Checkout button appears unresponsive during payment |
+| High confusion score on /pricing due to unclear CTA hierarchy | Users struggle to find the right pricing option     |
 
 Before completing the run, generate a concise title that summarizes what was fixed:
 
@@ -806,21 +977,29 @@ If the user agrees:
 2. Explain that recapt will monitor the affected pages/elements
 3. Suggest checking back in 24-48 hours with `evaluate_fix` to measure improvement
 
+## Tool Discovery
+
+Tools beyond the always-available set are dynamically registered. You must call `search_tools` once to activate a tool before calling it via `call_tool`. After the first search, call the tool directly for subsequent uses in the same phase.
+
+**Always available (no search needed):** `get_domains`, `run_full_diagnostic`, `triage_sessions`, `get_upgrade_options`, `search_tools`, `call_tool`, `memory_*`
+
 ## Tool Discovery Reference
 
-| Phase         | Search Query         | Tools                                                                                                   |
-| ------------- | -------------------- | ------------------------------------------------------------------------------------------------------- |
-| Run Tracking  | "improvement run"    | `start_improvement_run`, `update_improvement_run`, `record_improvement_action`, `list_improvement_runs` |
-| Check PRs     | "remediation status" | `list_remediations_by_status`, `check_mr_status`, `update_remediation_status`                           |
-| Check Pending | "pending fixes"      | `list_pending_fixes`, `evaluate_fix`                                                                    |
-| Diagnose      | (always available)   | `run_full_diagnostic`                                                                                   |
-| Journey       | "journey patterns"   | `get_journey_patterns`                                                                                  |
-| Funnels       | "analyze funnel"     | `analyze_funnel`                                                                                        |
-| Flows         | "analyze flow"       | `analyze_flow`, `get_flow_friction`                                                                     |
-| Personas      | "personas"           | `discover_personas`                                                                                     |
-| Compare       | "compare cohorts"    | `compare_cohorts`                                                                                       |
-| Investigate   | "investigate issue"  | `get_session_details`, `get_element_friction`, `get_page_metrics`, `triage_sessions`                    |
-| Audit         | "proposal"           | `create_proposal`, `list_proposals`, `evaluate_proposal`, `list_proposals_for_evaluation`               |
-| Fix           | "propose fix"        | `propose_fix`                                                                                           |
-| Track         | "deployment"         | `confirm_deployment`, `evaluate_fix`, `list_pending_fixes`                                              |
-| Learn         | "site knowledge"     | `get_site_knowledge`, `add_site_knowledge`                                                              |
+| Phase | Search Query | Tools |
+| ----- | ------------ | ----- |
+
+| Run Tracking | "improvement run" | `start_improvement_run`, `update_improvement_run`, `record_improvement_action`, `list_improvement_runs` |
+
+| Check PRs | "remediation status" | `list_remediations_by_status`, `check_mr_status`, `update_remediation_status` |
+| Check Pending | "pending fixes" | `list_pending_fixes`, `evaluate_fix` |
+| Diagnose | (always available) | `run_full_diagnostic` |
+| Journey | "journey patterns" | `get_journey_patterns` |
+| Funnels | "analyze funnel" | `analyze_funnel` |
+| Flows | "analyze flow" | `analyze_flow`, `get_flow_friction` |
+| Personas | "personas" | `discover_personas` |
+| Compare | "compare cohorts" | `compare_cohorts` |
+| Investigate | "investigate issue" | `get_session_details`, `get_element_friction`, `get_page_metrics`, `triage_sessions` |
+
+| Fix | "propose fix" | `propose_fix` |
+| Track | "deployment" | `confirm_deployment`, `evaluate_fix`, `list_pending_fixes` |
+| Learn | "site knowledge" | `get_site_knowledge`, `add_site_knowledge` |