@intentsolutionsio/skill-creator 5.0.0 → 5.0.3

package/skills/skill-creator/references/source-of-truth.md

@@ -3,6 +3,7 @@
 Canonical reference from [Anthropic docs](https://code.claude.com/docs/en/skills). Last synced: 2026-03-21.
 
 Additional references:
+
 - **Claude Code Extensions** — platform-specific fields ([changelog](https://code.claude.com/docs/en/changelog))
 - **Anthropic Engineering Blog** — progressive disclosure, degrees of freedom
 - **anthropics/skills** — [github.com/anthropics/skills](https://github.com/anthropics/skills) (official skill-creator reference implementation)
@@ -100,6 +101,7 @@ tags: [devops, automation]
 MCP tools use `ServerName:tool_name` format.
 
 Bash scoping patterns:
+
 ```yaml
 Bash(git:*)       # All git commands
 Bash(npm:*)       # All npm commands
@@ -141,6 +143,7 @@ Agents live in `agents/*.md` and use a different frontmatter schema than skills.
 ### Plugin Agent Restrictions
 
 When agents are distributed inside plugins, these fields are NOT supported:
+
 - `hooks`
 - `mcpServers`
 - `permissionMode`
@@ -222,17 +225,20 @@ skill-name/
 Skills use progressive disclosure to minimize context window usage.
 
 ### Level 1: Metadata (~100 tokens)
+
 - Frontmatter `name` and `description` only
 - Always loaded at startup for all installed skills
 - Aggregated into skill list in Claude's system prompt
 - **Budget**: ~2% of context window (configurable via `SLASH_COMMAND_TOOL_CHAR_BUDGET`)
 
 ### Level 2: SKILL.md Body (<5000 tokens / <500 lines)
+
 - Full instruction body loaded when skill activates
 - Contains workflow steps, examples, edge cases
 - Keep concise — Claude is already capable
 
 ### Level 3: Bundled Resources (unlimited)
+
 - `references/`, `scripts/`, `templates/`, `assets/`
 - Loaded only when explicitly needed during execution
 - Use clear section headers for navigability
@@ -307,7 +313,9 @@ description: "A helpful tool for documents"
 ## 6. Core Principles (Anthropic Official)
 
 ### Concise is Key
+
 Claude is already smart. Don't over-explain. Provide:
+
 - Clear workflow steps
 - Concrete examples
 - Edge cases that matter
@@ -343,6 +351,7 @@ Choose the right level. Over-constraining wastes tokens and fights Claude's capa
 ### Checklist Workflow Pattern
 
 For complex skills, structure the body as a checklist that Claude works through sequentially. Each item should be:
+
 - A concrete action (not a vague instruction)
 - Independently verifiable (Claude can confirm it's done)
 - Ordered by dependency (prerequisites first)
@@ -352,15 +361,17 @@ This pattern reduces skipped steps and improves consistency across models (Haiku
 ### Observation of Claude Navigation
 
 Claude navigates SKILL.md and references differently than humans:
+
 - **Reads top-down on first activation** — front-load the most important instructions
 - **Searches by heading** when returning to a section — use descriptive H2/H3 headers
-- **Follows markdown links eagerly** — a `[reference](./references/foo.md)` link will trigger a Read tool call
+- **Follows markdown links eagerly** — a `reference` link will trigger a Read tool call
 - **Skips content after long code blocks** — keep code examples short, move long ones to references
 - **Loses context in long files** — the 500-line limit exists because Claude's attention degrades past it
 
 ### Team Feedback
 
 When multiple authors maintain skills in a shared plugin:
+
 - Establish a shared glossary of terms used in descriptions (prevents synonym drift)
 - Use PR review checklists that include trigger-eval accuracy checks
 - Rotate skill ownership periodically to catch assumptions baked into instructions
@@ -369,16 +380,19 @@ When multiple authors maintain skills in a shared plugin:
 ### Description Optimization ("Pushy" Pattern)
 
 Skills frequently undertrigger because descriptions are too passive. Use aggressive claiming language:
+
 - "Make sure to use this skill whenever..." + specific scenarios
 - Front-load distinctive keywords
 - Include trigger phrases: "Use when...", "Activates for..."
 - Token budget: all descriptions load at startup (~15,000 char total via `SLASH_COMMAND_TOOL_CHAR_BUDGET`)
 
 ### No Time-Sensitive Information
+
 - Don't include dates, versions, or URLs that change
 - Reference tools by name, not version
 
 ### Consistent Terminology
+
 - Pick terms and stick with them throughout
 - Don't alternate between synonyms
 - Match terminology to the domain
@@ -445,7 +459,7 @@ Available in SKILL.md body for dynamic content:
 | `${CLAUDE_PLUGIN_ROOT}` | Hooks, plugin-level | Resolves to plugin root directory |
 | `${CLAUDE_PLUGIN_DATA}` | Persistent state | Survives updates/reinstalls (v2.1.78+) |
 
-Relative markdown links (`[API Reference](reference.md)`) work without path variables — Claude follows these with the Read tool on demand.
+Relative markdown links (`API Reference`) work without path variables — Claude follows these with the Read tool on demand.
 
 ### Usage Examples
 
@@ -488,59 +502,83 @@ Session tracking: ${CLAUDE_SESSION_ID}
 ## 9. Skill Patterns
 
 ### Script Automation
+
 Deterministic scripts that solve specific problems.
+
 ```
 skill activates -> runs script -> returns result
 ```
+
 Best for: file conversion, data transformation, API calls.
 
 ### Read-Process-Write
+
 Format conversion and transformation pipeline.
+
 ```
 read input -> process/transform -> write output
 ```
+
 Best for: document conversion, code generation, data formatting.
 
 ### Search-Analyze-Report
+
 Codebase analysis and reporting.
+
 ```
 search codebase -> analyze findings -> generate report
 ```
+
 Best for: code review, security audit, dependency analysis.
 
 ### Template-Based Generation
+
 Generate output from templates with variable substitution.
+
 ```
 load template -> fill variables -> validate -> output
 ```
+
 Best for: boilerplate generation, project scaffolding, config files.
 
 ### Wizard-Style Workflow
+
 Interactive multi-step gathering with AskUserQuestion.
+
 ```
 ask question -> gather input -> ask more -> generate result
 ```
+
 Best for: complex configuration, multi-option setup.
 
 ### Conditional Workflow
+
 Branch based on input or context.
+
 ```
 analyze input -> choose path -> execute branch -> output
 ```
+
 Best for: skills that handle multiple related tasks.
 
 ### Plan-Validate-Execute
+
 Verifiable intermediates with feedback loops.
+
 ```
 plan steps -> validate plan -> execute -> verify each step -> report
 ```
+
 Best for: deployment, migration, refactoring tasks.
 
 ### Visual Output Generation
+
 Generate HTML or visual artifacts.
+
 ```
 gather data -> generate HTML -> render preview
 ```
+
 Best for: dashboards, reports, documentation sites.
 
 ---

package/skills/skill-creator/references/validation-rules.md

@@ -1,4 +1,5 @@
 # Skill & Plugin Validation Rules
+
 Sources: [Anthropic docs](https://code.claude.com/docs/en/skills) · Intent Solutions enterprise policy
 
 Universal validation aligned with the Anthropic 2026 spec. Two tiers: Standard (Anthropic minimum) and Enterprise (our marketplace default — all fields required, zero tolerance for non-standard fields).
@@ -57,6 +58,7 @@ Body must contain all 7 sections (hard ERROR if any missing):
 ```
 
 Supporting files required (gold standard):
+
 - `PRD.md` must exist in skill root — Product Requirements Document
 - `ARD.md` must exist in skill root — Architecture Requirements Document
 - `references/` directory must exist (plural directory, NOT `reference.md` singular)
@@ -133,11 +135,13 @@ capabilities: []                  # NOTE: valid for agents ONLY, not skills
 ```
 
 **Plugin agents CANNOT use** (WARN if present):
+
 - `hooks` — plugin-level only, not agent-level
 - `mcpServers` — plugin-level only
 - `permissionMode` — standalone agent only, not plugin-scoped
 
 **Invalid for agents** (ERROR):
+
 - `expertise_level`, `activation_priority`, `color`, `activation_triggers`, `type`, `category` — invented, not Anthropic
 
 ---
@@ -237,6 +241,7 @@ Plus MCP tools in `ServerName:tool_name` format.
 | Enterprise | Error |
 
 Valid scoped patterns:
+
 ```
 Bash(git:*)
 Bash(npm:*)
@@ -284,6 +289,7 @@ Validate MCP server configuration structure.
 ### 7. Roll Up Plugin Score
 
 Plugin score = weighted average of component scores:
+
 - Skills: 50% weight
 - Agents: 20% weight
 - Commands: 15% weight
@@ -311,11 +317,13 @@ Anthropic defines 14 valid fields for agents. `name` and `description` are REQUI
 ### Context-Aware Rules
 
 **Plugin agents** (`plugins/*/agents/*.md`):
+
 - WARN if `hooks` present (hooks belong at plugin level, not agent level)
 - WARN if `mcpServers` present (plugin-level concern)
 - WARN if `permissionMode` present (standalone-only field)
 
 **Standalone agents** (`~/.claude/agents/*.md`):
+
 - All fields valid without restriction
 
 ### Invalid Agent Fields (ERROR)
@@ -409,6 +417,7 @@ The command runs at skill activation time. Output is injected verbatim into the
 ## String Substitution Validation
 
 If SKILL.md body contains `$ARGUMENTS` or `$0`, `$1`, etc.:
+
 - `argument-hint` SHOULD be set in frontmatter (WARNING if missing)
 - Instructions SHOULD handle empty `$ARGUMENTS` case
 - `$ARGUMENTS[N]` indexing should be sequential from 0
@@ -420,12 +429,14 @@ Also recognized: `${CLAUDE_SESSION_ID}` — current session identifier (Anthropi
 ## Validation Process
 
 ### Pre-flight
+
 1. File exists and is readable
 2. YAML frontmatter parses without error
 3. Frontmatter separator (`---`) present at start and end
 4. No non-standard fields present (ERROR on any invented/deprecated field)
 
 ### Field Validation
+
 1. All 8 required fields present (enterprise) or 2 required fields (standard)
 2. Field types correct (string, array, boolean, semver)
 3. Field constraints met (kebab-case, SPDX, valid tool names)
@@ -434,6 +445,7 @@ Also recognized: `${CLAUDE_SESSION_ID}` — current session identifier (Anthropi
 6. Conditional field logic (`context` requires `agent` and vice versa)
 
 ### Body Validation
+
 1. Length within limits (301-500 = WARNING, >500 = ERROR)
 2. All 7 required sections present (enterprise) — hard ERROR if any missing
 3. No absolute paths outside code blocks
@@ -442,15 +454,17 @@ Also recognized: `${CLAUDE_SESSION_ID}` — current session identifier (Anthropi
 6. `references/` directory exists (enterprise)
 
 ### Resource Validation
+
 1. All `${CLAUDE_SKILL_DIR}/scripts/*` references exist
 2. All `${CLAUDE_SKILL_DIR}/references/*` references exist
 3. All `${CLAUDE_SKILL_DIR}/templates/*` references exist
 4. All `${CLAUDE_SKILL_DIR}/assets/*` references exist
-5. Relative markdown links (e.g., `[ref](references/api.md)`) point to existing files
+5. Relative markdown links (e.g., `ref`) point to existing files
 6. No path escape attempts (`../`)
 7. No empty (0-byte) supporting files (stub detection)
 
 ### Report
+
 - Errors: Must fix (blocks pass)
 - Warnings: Should fix (does not block pass)
 - Info: Optional improvements (includes structural advisor suggestions)
@@ -465,21 +479,25 @@ Also recognized: `${CLAUDE_SESSION_ID}` — current session identifier (Anthropi
 INFO-level suggestions emitted after grading. Not scored — purely advisory.
 
 ### Split to Commands
+
 - **Trigger**: 3+ kebab-case `## operation-name` sections without `commands/` directory
 - **Suggestion**: Split into individual `commands/*.md` files
 - **Why**: Each operation becomes a separate slash command; skill stays lean
 
 ### Offload to References
+
 - **Trigger**: Body sections >20 lines (Output, Error Handling, Examples) without `references/`
 - **Suggestion**: Move to `references/section-name.md` with relative markdown link
 - **Why**: Reduces token footprint; Claude reads on demand
 
 ### DCI Opportunities
+
 - **Trigger**: File existence checks, git operations, or tool version detection without DCI
 - **Suggestion**: Add `` !`command` `` directives for auto-detection at activation
 - **Why**: Eliminates discovery tool calls; Claude starts with context pre-loaded
 
 ### Migrate Commands to Skills
+
 - **Trigger**: `commands/*.md` files present without corresponding `skills/` entries
 - **Suggestion**: Consider migrating to SKILL.md format for auto-activation
 - **Why**: Skills activate automatically on context; commands require explicit `/name` invocation

package/skills/skill-creator/scripts/aggregate_benchmark.py

@@ -60,7 +60,7 @@ def calculate_stats(values: list[float]) -> dict:
         "mean": round(mean, 4),
         "stddev": round(stddev, 4),
         "min": round(min(values), 4),
-        "max": round(max(values), 4)
+        "max": round(max(values), 4),
     }
 
 
@@ -157,7 +157,9 @@ def load_run_results(benchmark_dir: Path) -> dict:
                 raw_expectations = grading.get("expectations", [])
                 for exp in raw_expectations:
                     if "text" not in exp or "passed" not in exp:
-                        print(f"Warning: expectation in {grading_file} missing required fields (text, passed, evidence): {exp}")
+                        print(
+                            f"Warning: expectation in {grading_file} missing required fields (text, passed, evidence): {exp}"
+                        )
                 result["expectations"] = raw_expectations
 
                 # Extract notes from user_notes_summary
@@ -189,7 +191,7 @@ def aggregate_results(results: dict) -> dict:
             run_summary[config] = {
                 "pass_rate": {"mean": 0.0, "stddev": 0.0, "min": 0.0, "max": 0.0},
                 "time_seconds": {"mean": 0.0, "stddev": 0.0, "min": 0.0, "max": 0.0},
-                "tokens": {"mean": 0, "stddev": 0, "min": 0, "max": 0}
+                "tokens": {"mean": 0, "stddev": 0, "min": 0, "max": 0},
             }
             continue
 
@@ -200,7 +202,7 @@ def aggregate_results(results: dict) -> dict:
         run_summary[config] = {
             "pass_rate": calculate_stats(pass_rates),
             "time_seconds": calculate_stats(times),
-            "tokens": calculate_stats(tokens)
+            "tokens": calculate_stats(tokens),
         }
 
     # Calculate delta between the first two configs (if two exist)
@@ -218,7 +220,7 @@ def aggregate_results(results: dict) -> dict:
     run_summary["delta"] = {
         "pass_rate": f"{delta_pass_rate:+.2f}",
         "time_seconds": f"{delta_time:+.1f}",
-        "tokens": f"{delta_tokens:+.0f}"
+        "tokens": f"{delta_tokens:+.0f}",
     }
 
     return run_summary
@@ -235,30 +237,28 @@ def generate_benchmark(benchmark_dir: Path, skill_name: str = "", skill_path: st
     runs = []
     for config in results:
         for result in results[config]:
-            runs.append({
-                "eval_id": result["eval_id"],
-                "configuration": config,
-                "run_number": result["run_number"],
-                "result": {
-                    "pass_rate": result["pass_rate"],
-                    "passed": result["passed"],
-                    "failed": result["failed"],
-                    "total": result["total"],
-                    "time_seconds": result["time_seconds"],
-                    "tokens": result.get("tokens", 0),
-                    "tool_calls": result.get("tool_calls", 0),
-                    "errors": result.get("errors", 0)
-                },
-                "expectations": result["expectations"],
-                "notes": result["notes"]
-            })
+            runs.append(
+                {
+                    "eval_id": result["eval_id"],
+                    "configuration": config,
+                    "run_number": result["run_number"],
+                    "result": {
+                        "pass_rate": result["pass_rate"],
+                        "passed": result["passed"],
+                        "failed": result["failed"],
+                        "total": result["total"],
+                        "time_seconds": result["time_seconds"],
+                        "tokens": result.get("tokens", 0),
+                        "tool_calls": result.get("tool_calls", 0),
+                        "errors": result.get("errors", 0),
+                    },
+                    "expectations": result["expectations"],
+                    "notes": result["notes"],
+                }
+            )
 
     # Determine eval IDs from results
-    eval_ids = sorted(set(
-        r["eval_id"]
-        for config in results.values()
-        for r in config
-    ))
+    eval_ids = sorted(set(r["eval_id"] for config in results.values() for r in config))
 
     benchmark = {
         "metadata": {
@@ -268,11 +268,11 @@ def generate_benchmark(benchmark_dir: Path, skill_name: str = "", skill_path: st
             "analyzer_model": "<model-name>",
             "timestamp": datetime.now(timezone.utc).strftime("%Y-%m-%dT%H:%M:%SZ"),
             "evals_run": eval_ids,
-            "runs_per_configuration": 3
+            "runs_per_configuration": 3,
         },
         "runs": runs,
         "run_summary": run_summary,
-        "notes": []  # To be filled by analyzer
+        "notes": [],  # To be filled by analyzer
     }
 
     return benchmark
@@ -310,25 +310,27 @@ def generate_markdown(benchmark: dict) -> str:
     # Format pass rate
     a_pr = a_summary.get("pass_rate", {})
     b_pr = b_summary.get("pass_rate", {})
-    lines.append(f"| Pass Rate | {a_pr.get('mean', 0)*100:.0f}% ± {a_pr.get('stddev', 0)*100:.0f}% | {b_pr.get('mean', 0)*100:.0f}% ± {b_pr.get('stddev', 0)*100:.0f}% | {delta.get('pass_rate', '—')} |")
+    lines.append(
+        f"| Pass Rate | {a_pr.get('mean', 0) * 100:.0f}% ± {a_pr.get('stddev', 0) * 100:.0f}% | {b_pr.get('mean', 0) * 100:.0f}% ± {b_pr.get('stddev', 0) * 100:.0f}% | {delta.get('pass_rate', '—')} |"
+    )
 
     # Format time
     a_time = a_summary.get("time_seconds", {})
     b_time = b_summary.get("time_seconds", {})
-    lines.append(f"| Time | {a_time.get('mean', 0):.1f}s ± {a_time.get('stddev', 0):.1f}s | {b_time.get('mean', 0):.1f}s ± {b_time.get('stddev', 0):.1f}s | {delta.get('time_seconds', '—')}s |")
+    lines.append(
+        f"| Time | {a_time.get('mean', 0):.1f}s ± {a_time.get('stddev', 0):.1f}s | {b_time.get('mean', 0):.1f}s ± {b_time.get('stddev', 0):.1f}s | {delta.get('time_seconds', '—')}s |"
+    )
 
     # Format tokens
     a_tokens = a_summary.get("tokens", {})
     b_tokens = b_summary.get("tokens", {})
-    lines.append(f"| Tokens | {a_tokens.get('mean', 0):.0f} ± {a_tokens.get('stddev', 0):.0f} | {b_tokens.get('mean', 0):.0f} ± {b_tokens.get('stddev', 0):.0f} | {delta.get('tokens', '—')} |")
+    lines.append(
+        f"| Tokens | {a_tokens.get('mean', 0):.0f} ± {a_tokens.get('stddev', 0):.0f} | {b_tokens.get('mean', 0):.0f} ± {b_tokens.get('stddev', 0):.0f} | {delta.get('tokens', '—')} |"
+    )
 
     # Notes section
     if benchmark.get("notes"):
-        lines.extend([
-            "",
-            "## Notes",
-            ""
-        ])
+        lines.extend(["", "## Notes", ""])
         for note in benchmark["notes"]:
             lines.append(f"- {note}")
 
@@ -336,28 +338,12 @@ def generate_markdown(benchmark: dict) -> str:
 
 
 def main():
-    parser = argparse.ArgumentParser(
-        description="Aggregate benchmark run results into summary statistics"
-    )
-    parser.add_argument(
-        "benchmark_dir",
-        type=Path,
-        help="Path to the benchmark directory"
-    )
-    parser.add_argument(
-        "--skill-name",
-        default="",
-        help="Name of the skill being benchmarked"
-    )
-    parser.add_argument(
-        "--skill-path",
-        default="",
-        help="Path to the skill being benchmarked"
-    )
+    parser = argparse.ArgumentParser(description="Aggregate benchmark run results into summary statistics")
+    parser.add_argument("benchmark_dir", type=Path, help="Path to the benchmark directory")
+    parser.add_argument("--skill-name", default="", help="Name of the skill being benchmarked")
+    parser.add_argument("--skill-path", default="", help="Path to the skill being benchmarked")
     parser.add_argument(
-        "--output", "-o",
-        type=Path,
-        help="Output path for benchmark.json (default: <benchmark_dir>/benchmark.json)"
+        "--output", "-o", type=Path, help="Output path for benchmark.json (default: <benchmark_dir>/benchmark.json)"
     )
 
     args = parser.parse_args()
@@ -389,11 +375,11 @@ def main():
     configs = [k for k in run_summary if k != "delta"]
     delta = run_summary.get("delta", {})
 
-    print(f"\nSummary:")
+    print("\nSummary:")
     for config in configs:
         pr = run_summary[config]["pass_rate"]["mean"]
         label = config.replace("_", " ").title()
-        print(f"  {label}: {pr*100:.1f}% pass rate")
+        print(f"  {label}: {pr * 100:.1f}% pass rate")
     print(f"  Delta:         {delta.get('pass_rate', '—')}")
 
 

package/skills/skill-creator/scripts/generate_report.py

@@ -16,7 +16,7 @@ from pathlib import Path
 def generate_html(data: dict, auto_refresh: bool = False, skill_name: str = "") -> str:
     """Generate HTML report from loop output data. If auto_refresh is True, adds a meta refresh tag."""
     history = data.get("history", [])
-    holdout = data.get("holdout", 0)
+    data.get("holdout", 0)
     title_prefix = html.escape(skill_name + " \u2014 ") if skill_name else ""
 
     # Get all unique queries from train and test sets, with should_trigger info
@@ -31,11 +31,16 @@ def generate_html(data: dict, auto_refresh: bool = False, skill_name: str = "")
 
     refresh_tag = '    <meta http-equiv="refresh" content="5">\n' if auto_refresh else ""
 
-    html_parts = ["""<!DOCTYPE html>
+    html_parts = [
+        """<!DOCTYPE html>
 <html>
 <head>
     <meta charset="utf-8">
-""" + refresh_tag + """    <title>""" + title_prefix + """Skill Description Optimization</title>
+"""
+        + refresh_tag
+        + """    <title>"""
+        + title_prefix
+        + """Skill Description Optimization</title>
     <link rel="preconnect" href="https://fonts.googleapis.com">
     <link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
     <link href="https://fonts.googleapis.com/css2?family=Poppins:wght@500;600&family=Lora:wght@400;500&display=swap" rel="stylesheet">
@@ -146,21 +151,24 @@ def generate_html(data: dict, auto_refresh: bool = False, skill_name: str = "")
     </style>
 </head>
 <body>
-    <h1>""" + title_prefix + """Skill Description Optimization</h1>
+    <h1>"""
+        + title_prefix
+        + """Skill Description Optimization</h1>
     <div class="explainer">
         <strong>Optimizing your skill's description.</strong> This page updates automatically as Claude tests different versions of your skill's description. Each row is an iteration — a new description attempt. The columns show test queries: green checkmarks mean the skill triggered correctly (or correctly didn't trigger), red crosses mean it got it wrong. The "Train" score shows performance on queries used to improve the description; the "Test" score shows performance on held-out queries the optimizer hasn't seen. When it's done, Claude will apply the best-performing description to your skill.
     </div>
-"""]
+"""
+    ]
 
     # Summary section
-    best_test_score = data.get('best_test_score')
-    best_train_score = data.get('best_train_score')
+    best_test_score = data.get("best_test_score")
+    data.get("best_train_score")
     html_parts.append(f"""
     <div class="summary">
-        <p><strong>Original:</strong> {html.escape(data.get('original_description', 'N/A'))}</p>
-        <p class="best"><strong>Best:</strong> {html.escape(data.get('best_description', 'N/A'))}</p>
-        <p><strong>Best Score:</strong> {data.get('best_score', 'N/A')} {'(test)' if best_test_score else '(train)'}</p>
-        <p><strong>Iterations:</strong> {data.get('iterations_run', 0)} | <strong>Train:</strong> {data.get('train_size', '?')} | <strong>Test:</strong> {data.get('test_size', '?')}</p>
+        <p><strong>Original:</strong> {html.escape(data.get("original_description", "N/A"))}</p>
+        <p class="best"><strong>Best:</strong> {html.escape(data.get("best_description", "N/A"))}</p>
+        <p><strong>Best Score:</strong> {data.get("best_score", "N/A")} {"(test)" if best_test_score else "(train)"}</p>
+        <p><strong>Iterations:</strong> {data.get("iterations_run", 0)} | <strong>Train:</strong> {data.get("train_size", "?")} | <strong>Test:</strong> {data.get("test_size", "?")}</p>
     </div>
 """)
 
@@ -211,10 +219,10 @@ def generate_html(data: dict, auto_refresh: bool = False, skill_name: str = "")
     # Add rows for each iteration
     for h in history:
         iteration = h.get("iteration", "?")
-        train_passed = h.get("train_passed", h.get("passed", 0))
-        train_total = h.get("train_total", h.get("total", 0))
-        test_passed = h.get("test_passed")
-        test_total = h.get("test_total")
+        h.get("train_passed", h.get("passed", 0))
+        h.get("train_total", h.get("total", 0))
+        h.get("test_passed")
+        h.get("test_total")
         description = h.get("description", "")
         train_results = h.get("train_results", h.get("results", []))
         test_results = h.get("test_results", [])
@@ -272,7 +280,9 @@ def generate_html(data: dict, auto_refresh: bool = False, skill_name: str = "")
             icon = "✓" if did_pass else "✗"
             css_class = "pass" if did_pass else "fail"
 
-            html_parts.append(f'                <td class="result {css_class}">{icon}<span class="rate">{triggers}/{runs}</span></td>\n')
+            html_parts.append(
+                f'                <td class="result {css_class}">{icon}<span class="rate">{triggers}/{runs}</span></td>\n'
+            )
 
         # Add result for each test query (with different background)
         for qinfo in test_queries:
@@ -284,7 +294,9 @@ def generate_html(data: dict, auto_refresh: bool = False, skill_name: str = "")
             icon = "✓" if did_pass else "✗"
             css_class = "pass" if did_pass else "fail"
 
-            html_parts.append(f'                <td class="result test-result {css_class}">{icon}<span class="rate">{triggers}/{runs}</span></td>\n')
+            html_parts.append(
+                f'                <td class="result test-result {css_class}">{icon}<span class="rate">{triggers}/{runs}</span></td>\n'
+            )
 
         html_parts.append("            </tr>\n")