@rotifer/playground 0.7.9 → 0.8.0

package/genes/guard-strict/phenotype.json

@@ -0,0 +1,107 @@
+{
+  "domain": "guard.code-review",
+  "description": "Strict guard gene for code review output filtering. High precision, low recall — blocks false positives aggressively. Prefers missing a real issue over flagging a non-issue. Part of the Code Review Assistant 3×2 arena competition.",
+  "inputSchema": {
+    "type": "object",
+    "properties": {
+      "reviewOutput": {
+        "type": "object",
+        "description": "Raw output from a Prompt Gene (code review findings)",
+        "properties": {
+          "findings": {
+            "type": "array",
+            "items": {
+              "type": "object",
+              "properties": {
+                "severity": { "type": "string" },
+                "category": { "type": "string" },
+                "message": { "type": "string" },
+                "line": { "type": "integer" },
+                "fix": { "type": "string" }
+              }
+            }
+          },
+          "summary": { "type": "string" }
+        }
+      },
+      "originalCode": {
+        "type": "string",
+        "description": "The original code being reviewed (for cross-checking)"
+      }
+    },
+    "required": ["reviewOutput"]
+  },
+  "outputSchema": {
+    "type": "object",
+    "properties": {
+      "filteredFindings": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "severity": { "type": "string" },
+            "category": { "type": "string" },
+            "message": { "type": "string" },
+            "line": { "type": "integer" },
+            "fix": { "type": "string" },
+            "confidence": { "type": "number", "description": "0-1 confidence that this is a real issue" }
+          }
+        },
+        "description": "Only findings that pass strict validation"
+      },
+      "rejected": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "originalMessage": { "type": "string" },
+            "rejectReason": { "type": "string" }
+          }
+        },
+        "description": "Findings rejected with reason"
+      },
+      "guardStats": {
+        "type": "object",
+        "properties": {
+          "inputCount": { "type": "integer" },
+          "outputCount": { "type": "integer" },
+          "rejectRate": { "type": "number" }
+        }
+      }
+    }
+  },
+  "llmRequirements": {
+    "templateFormat": "mustache",
+    "templateVariables": ["reviewOutput", "originalCode"],
+    "targetModels": ["claude-3", "gpt-4"],
+    "minContextWindow": 8192,
+    "expectedOutputFormat": "json",
+    "temperatureHint": 0.0,
+    "maxOutputTokens": 4096,
+    "systemPromptPath": "system-prompt.md",
+    "chainOfThought": true
+  },
+  "guardConfig": {
+    "mode": "filter",
+    "vgContribution": "Security_Leak_Risk",
+    "strictness": "high",
+    "targetFalsePositiveRate": 0.05,
+    "targetFalseNegativeRate": 0.30
+  },
+  "fitnessConfig": {
+    "evaluationMethod": "template-quality",
+    "dimensions": {
+      "precision": { "weight": 0.40, "description": "1 - false positive rate" },
+      "recall": { "weight": 0.20, "description": "Fraction of true issues preserved" },
+      "rejectQuality": { "weight": 0.20, "description": "Correctness of reject reasons" },
+      "templateValidity": { "weight": 0.10, "description": "Template renders without error" },
+      "resourceEfficiency": { "weight": 0.10, "description": "Token efficiency" }
+    }
+  },
+  "dependencies": [],
+  "version": "0.1.0",
+  "author": "rotifer-team",
+  "fidelity": "Wrapped",
+  "transparency": "Open",
+  "source": "development-genome"
+}

package/genes/guard-strict/system-prompt.md

@@ -0,0 +1,18 @@
+You are a STRICT code review guard. Your job is to filter code review findings, removing likely false positives.
+
+Your philosophy: **Better to miss a real issue than to waste developer time on a false alarm.**
+
+For each finding in the review output, evaluate:
+
+1. **Line reference check**: Does the finding reference a specific line? If the line number doesn't correspond to relevant code, REJECT.
+2. **Category validity**: Is the category (e.g., "injection", "memory-leak") supported by the actual code pattern? Reject vague or generic warnings.
+3. **Fix actionability**: Does the fix suggestion contain concrete code? Reject findings with only "consider reviewing" or "be careful" advice.
+4. **Hallucination check**: If `originalCode` is provided, verify the finding references actual code constructs. Reject if the finding describes code that doesn't exist.
+5. **Severity calibration**: Downgrade "critical" to "warning" if the exploit requires unlikely preconditions.
+
+Only pass findings with confidence >= 0.7.
+
+Output valid JSON with:
+- `filteredFindings`: findings that survived filtering (with confidence scores)
+- `rejected`: findings removed (with reject reasons for transparency)
+- `guardStats`: input/output counts and reject rate

package/genes/prompt-review-perf/.gene-manifest.json

@@ -0,0 +1,8 @@
+{
+  "geneId": "41b58d584a54b17faa96829c39b5e6e4c5cb71af2adfba6ac670bb96ae8c9bba",
+  "name": "prompt-review-perf",
+  "domain": "prompt.code-review",
+  "fidelity": "Wrapped",
+  "wrappedAt": "2026-03-24T01:54:54.000Z",
+  "fromSkill": null
+}

package/genes/prompt-review-perf/phenotype.json

@@ -0,0 +1,61 @@
+{
+  "domain": "prompt.code-review",
+  "description": "Code review gene specialized in performance issue detection. Identifies N+1 queries, unnecessary re-renders, memory leaks, blocking operations, inefficient algorithms, and bundle size issues. Part of the Code Review Assistant 3×2 arena competition.",
+  "inputSchema": {
+    "type": "object",
+    "properties": {
+      "code": { "type": "string", "description": "Source code to review" },
+      "language": { "type": "string", "description": "Programming language" },
+      "context": { "type": "string", "description": "File path or component context" }
+    },
+    "required": ["code"]
+  },
+  "outputSchema": {
+    "type": "object",
+    "properties": {
+      "findings": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "severity": { "type": "string", "enum": ["critical", "warning", "info"] },
+            "category": { "type": "string" },
+            "message": { "type": "string" },
+            "line": { "type": "integer" },
+            "fix": { "type": "string" },
+            "impact": { "type": "string", "description": "Estimated performance impact" }
+          }
+        }
+      },
+      "summary": { "type": "string" },
+      "overallImpact": { "type": "string", "enum": ["high", "medium", "low", "negligible"] }
+    }
+  },
+  "llmRequirements": {
+    "templateFormat": "mustache",
+    "templateVariables": ["code", "language", "context"],
+    "targetModels": ["claude-3", "gpt-4"],
+    "minContextWindow": 8192,
+    "expectedOutputFormat": "json",
+    "temperatureHint": 0.2,
+    "maxOutputTokens": 4096,
+    "systemPromptPath": "system-prompt.md",
+    "chainOfThought": true
+  },
+  "fitnessConfig": {
+    "evaluationMethod": "template-quality",
+    "dimensions": {
+      "detectionRate": { "weight": 0.35, "description": "Fraction of known perf issues found" },
+      "falsePositiveRate": { "weight": 0.25, "description": "Precision of findings" },
+      "responseQuality": { "weight": 0.20, "description": "Actionability of optimization suggestions" },
+      "templateValidity": { "weight": 0.10, "description": "Template renders without error" },
+      "resourceEfficiency": { "weight": 0.10, "description": "Token efficiency" }
+    }
+  },
+  "dependencies": [],
+  "version": "0.1.0",
+  "author": "rotifer-team",
+  "fidelity": "Wrapped",
+  "transparency": "Open",
+  "source": "development-genome"
+}

package/genes/prompt-review-perf/system-prompt.md

@@ -0,0 +1,22 @@
+You are a performance-focused code reviewer. Your mission is to find performance bottlenecks and optimization opportunities.
+
+Focus areas (in priority order):
+1. **Database queries**: N+1 queries, missing indexes, unbounded SELECTs, unnecessary JOINs
+2. **Memory**: leaks, unbounded caches, large object retention, missing cleanup
+3. **Async/blocking**: synchronous operations in async context, missing concurrency, sequential awaits that could be parallel
+4. **Rendering**: unnecessary re-renders (React), layout thrashing, unoptimized images
+5. **Algorithms**: O(n²) when O(n log n) exists, redundant computation, missing memoization
+6. **Bundle/payload**: unused imports, large dependencies for small tasks, missing tree-shaking
+7. **Network**: missing caching headers, redundant API calls, large payloads
+
+For each finding, provide:
+- `severity`: "critical" (measurable impact >100ms), "warning" (noticeable), "info" (micro-optimization)
+- `category`: short tag (e.g., "n-plus-1", "memory-leak", "blocking-io")
+- `message`: clear description of the issue
+- `line`: approximate line number
+- `fix`: concrete optimization with expected improvement
+- `impact`: estimated performance impact (e.g., "~200ms savings per request")
+
+Focus on findings with measurable impact. Avoid premature optimization suggestions.
+
+Output valid JSON matching the output schema.

package/genes/prompt-review-readability/.gene-manifest.json

@@ -0,0 +1,8 @@
+{
+  "geneId": "ff84e0a13591b78c08dce76c0998b60c725958380b45459ad933cd211f7dae68",
+  "name": "prompt-review-readability",
+  "domain": "prompt.code-review",
+  "fidelity": "Wrapped",
+  "wrappedAt": "2026-03-24T01:54:54.000Z",
+  "fromSkill": null
+}

package/genes/prompt-review-readability/phenotype.json

@@ -0,0 +1,60 @@
+{
+  "domain": "prompt.code-review",
+  "description": "Code review gene specialized in readability and maintainability. Identifies unclear naming, missing documentation, excessive complexity, inconsistent patterns, and code smell. Part of the Code Review Assistant 3×2 arena competition.",
+  "inputSchema": {
+    "type": "object",
+    "properties": {
+      "code": { "type": "string", "description": "Source code to review" },
+      "language": { "type": "string", "description": "Programming language" },
+      "context": { "type": "string", "description": "File path or component context" }
+    },
+    "required": ["code"]
+  },
+  "outputSchema": {
+    "type": "object",
+    "properties": {
+      "findings": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "severity": { "type": "string", "enum": ["critical", "warning", "info"] },
+            "category": { "type": "string" },
+            "message": { "type": "string" },
+            "line": { "type": "integer" },
+            "fix": { "type": "string" }
+          }
+        }
+      },
+      "summary": { "type": "string" },
+      "readabilityScore": { "type": "number", "description": "0-10 readability score" }
+    }
+  },
+  "llmRequirements": {
+    "templateFormat": "mustache",
+    "templateVariables": ["code", "language", "context"],
+    "targetModels": ["claude-3", "gpt-4"],
+    "minContextWindow": 8192,
+    "expectedOutputFormat": "json",
+    "temperatureHint": 0.3,
+    "maxOutputTokens": 4096,
+    "systemPromptPath": "system-prompt.md",
+    "chainOfThought": false
+  },
+  "fitnessConfig": {
+    "evaluationMethod": "template-quality",
+    "dimensions": {
+      "detectionRate": { "weight": 0.30, "description": "Coverage of known readability issues" },
+      "falsePositiveRate": { "weight": 0.25, "description": "Precision — avoid pedantic nitpicks" },
+      "responseQuality": { "weight": 0.25, "description": "Quality of rewrite suggestions" },
+      "templateValidity": { "weight": 0.10, "description": "Template renders without error" },
+      "resourceEfficiency": { "weight": 0.10, "description": "Token efficiency" }
+    }
+  },
+  "dependencies": [],
+  "version": "0.1.0",
+  "author": "rotifer-team",
+  "fidelity": "Wrapped",
+  "transparency": "Open",
+  "source": "development-genome"
+}

package/genes/prompt-review-readability/system-prompt.md

@@ -0,0 +1,21 @@
+You are a readability-focused code reviewer. Your mission is to improve code clarity and maintainability.
+
+Focus areas (in priority order):
+1. **Naming**: unclear variable/function/class names, abbreviations, misleading names
+2. **Complexity**: functions >30 lines, nesting >3 levels, cyclomatic complexity >10
+3. **Documentation**: missing JSDoc/docstrings for public APIs, outdated comments
+4. **Patterns**: inconsistent error handling, mixed paradigms, anti-patterns
+5. **Structure**: god functions, missing abstractions, tight coupling
+6. **Dead code**: unused variables, unreachable branches, commented-out code
+7. **Consistency**: mixed naming conventions, inconsistent formatting within file
+
+For each finding, provide:
+- `severity`: "critical" (blocks understanding), "warning" (slows comprehension), "info" (style preference)
+- `category`: short tag (e.g., "naming", "complexity", "dead-code")
+- `message`: clear description of the readability issue
+- `line`: approximate line number
+- `fix`: concrete rewrite suggestion (show before/after when possible)
+
+Respect the codebase's existing style. Don't enforce personal preferences — focus on objective clarity improvements.
+
+Output valid JSON matching the output schema. Include a `readabilityScore` from 0 (unreadable) to 10 (exemplary).

package/genes/prompt-review-security/.gene-manifest.json

@@ -0,0 +1,8 @@
+{
+  "geneId": "0376427b53f9d478290a5908646720cb2772cca42981c85a84448cfa6eefca00",
+  "name": "prompt-review-security",
+  "domain": "prompt.code-review",
+  "fidelity": "Wrapped",
+  "wrappedAt": "2026-03-24T01:54:54.000Z",
+  "fromSkill": null
+}

package/genes/prompt-review-security/phenotype.json

@@ -0,0 +1,60 @@
+{
+  "domain": "prompt.code-review",
+  "description": "Code review gene specialized in security vulnerability detection. Identifies prompt injection, private key exposure, path traversal, SSRF, input validation gaps, and supply chain risks. Part of the Code Review Assistant 3×2 arena competition.",
+  "inputSchema": {
+    "type": "object",
+    "properties": {
+      "code": { "type": "string", "description": "Source code to review" },
+      "language": { "type": "string", "description": "Programming language" },
+      "context": { "type": "string", "description": "File path or component context" }
+    },
+    "required": ["code"]
+  },
+  "outputSchema": {
+    "type": "object",
+    "properties": {
+      "findings": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "severity": { "type": "string", "enum": ["critical", "warning", "info"] },
+            "category": { "type": "string" },
+            "message": { "type": "string" },
+            "line": { "type": "integer" },
+            "fix": { "type": "string" }
+          }
+        }
+      },
+      "summary": { "type": "string" },
+      "riskLevel": { "type": "string", "enum": ["high", "medium", "low", "none"] }
+    }
+  },
+  "llmRequirements": {
+    "templateFormat": "mustache",
+    "templateVariables": ["code", "language", "context"],
+    "targetModels": ["claude-3", "gpt-4"],
+    "minContextWindow": 8192,
+    "expectedOutputFormat": "json",
+    "temperatureHint": 0.1,
+    "maxOutputTokens": 4096,
+    "systemPromptPath": "system-prompt.md",
+    "chainOfThought": true
+  },
+  "fitnessConfig": {
+    "evaluationMethod": "template-quality",
+    "dimensions": {
+      "detectionRate": { "weight": 0.35, "description": "Fraction of known issues found in test corpus" },
+      "falsePositiveRate": { "weight": 0.25, "description": "1 - (false positives / total findings), higher is better" },
+      "responseQuality": { "weight": 0.20, "description": "Actionability of fix suggestions" },
+      "templateValidity": { "weight": 0.10, "description": "Template renders without error" },
+      "resourceEfficiency": { "weight": 0.10, "description": "Token efficiency of rendered prompt" }
+    }
+  },
+  "dependencies": [],
+  "version": "0.1.0",
+  "author": "rotifer-team",
+  "fidelity": "Wrapped",
+  "transparency": "Open",
+  "source": "development-genome"
+}

package/genes/prompt-review-security/system-prompt.md

@@ -0,0 +1,21 @@
+You are a security-focused code reviewer. Your mission is to find security vulnerabilities in the provided code.
+
+Focus areas (in priority order):
+1. **Injection attacks**: SQL injection, prompt injection, command injection, XSS
+2. **Secret exposure**: hardcoded API keys, private keys, credentials in source
+3. **Path traversal**: `../` in file paths, unsanitized user input in file operations
+4. **SSRF/DNS rebinding**: fetch to user-controlled URLs, private IP access
+5. **Input validation**: missing type checks, unbounded inputs, prototype pollution
+6. **Authentication/Authorization**: missing auth checks, broken access control
+7. **Supply chain**: known vulnerable dependencies, typosquatting risks
+
+For each finding, provide:
+- `severity`: "critical" (exploitable), "warning" (potential risk), "info" (best practice)
+- `category`: short tag (e.g., "injection", "secret-exposure", "path-traversal")
+- `message`: clear description of the vulnerability
+- `line`: approximate line number
+- `fix`: concrete fix suggestion
+
+Be thorough but precise — false positives erode trust. Only flag genuine risks.
+
+Output valid JSON matching the output schema.

package/genes/rotifer-protocol/SKILL.md

@@ -117,5 +117,5 @@ rotifer agent run search-bot --input '{"query":"hello"}'
 - [rotifer.dev](https://rotifer.dev) — Website
 - [rotifer.dev/genes](https://rotifer.dev/genes/) — Gene Store
 - [rotifer.dev/docs](https://rotifer.dev/docs/) — Documentation
-- [GitLab](https://gitlab.com/rotifer-protocol/rotifer-playground) — Source
+- [GitHub](https://github.com/rotifer-protocol/rotifer-playground) — Source
 - [Discord](https://discord.gg/6d4JrfMr) — Community

package/genes/rule-router-frequency/.gene-manifest.json

@@ -0,0 +1,8 @@
+{
+  "geneId": "4e172b696388f8a3d488241dba2a62360567c4add7bca43aa93b3ec6184dcee8",
+  "name": "rule-router-frequency",
+  "domain": "prompt.rule-router",
+  "fidelity": "Wrapped",
+  "wrappedAt": "2026-03-24T01:54:52.000Z",
+  "fromSkill": null
+}

package/genes/rule-router-frequency/phenotype.json

@@ -0,0 +1,76 @@
+{
+  "domain": "prompt.rule-router",
+  "description": "Rule routing strategy that prioritizes high-frequency rules. Ranks .cursor/rules by historical trigger frequency, injecting the most-triggered rules first to maximize context hit rate. Variant A of the Rule Router Gene competition.",
+  "inputSchema": {
+    "type": "object",
+    "properties": {
+      "userMessage": {
+        "type": "string",
+        "description": "The user's current message/query"
+      },
+      "availableRules": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "name": { "type": "string" },
+            "triggerWords": { "type": "array", "items": { "type": "string" } },
+            "frequency": { "type": "integer", "description": "Historical trigger count" },
+            "sizeTokens": { "type": "integer" }
+          }
+        },
+        "description": "Available rules with metadata"
+      },
+      "contextBudget": {
+        "type": "integer",
+        "description": "Maximum tokens available for rule injection",
+        "default": 4096
+      }
+    },
+    "required": ["userMessage", "availableRules"]
+  },
+  "outputSchema": {
+    "type": "object",
+    "properties": {
+      "selectedRules": {
+        "type": "array",
+        "items": { "type": "string" },
+        "description": "Ordered list of rule names to inject"
+      },
+      "totalTokens": {
+        "type": "integer",
+        "description": "Total token cost of selected rules"
+      },
+      "strategy": {
+        "type": "string",
+        "description": "Routing strategy explanation"
+      }
+    }
+  },
+  "llmRequirements": {
+    "templateFormat": "mustache",
+    "templateVariables": ["userMessage", "availableRules", "contextBudget"],
+    "targetModels": ["claude-3", "gpt-4"],
+    "minContextWindow": 4096,
+    "expectedOutputFormat": "json",
+    "temperatureHint": 0.1,
+    "maxOutputTokens": 1024,
+    "systemPromptPath": "system-prompt.md",
+    "chainOfThought": false
+  },
+  "fitnessConfig": {
+    "evaluationMethod": "template-quality",
+    "dimensions": {
+      "routeHitRate": { "weight": 0.40, "description": "Fraction of injected rules actually triggered by user message" },
+      "contextSavings": { "weight": 0.30, "description": "1 - (usedTokens / contextBudget)" },
+      "templateValidity": { "weight": 0.20, "description": "Template renders without error for all valid inputs" },
+      "crossModelConsistency": { "weight": 0.10, "description": "Same input produces same rule selection across models" }
+    }
+  },
+  "dependencies": [],
+  "version": "0.1.0",
+  "author": "rotifer-team",
+  "fidelity": "Wrapped",
+  "transparency": "Open",
+  "source": "development-genome"
+}

package/genes/rule-router-frequency/system-prompt.md

@@ -0,0 +1,26 @@
+You are a Rule Router that selects which .cursor/rules to inject into the AI context.
+
+Your strategy: **FREQUENCY-FIRST ROUTING**
+
+Given a user message and a list of available rules, select rules by prioritizing those with the highest historical trigger frequency. This maximizes the probability that injected rules will be relevant, based on past usage patterns.
+
+Algorithm:
+1. Sort available rules by `frequency` (descending)
+2. For each rule (in frequency order), check if any `triggerWords` appear in the user message
+3. If a trigger word matches, add the rule to the selection
+4. Continue until `contextBudget` is exhausted or all matching rules are selected
+5. If budget remains after all matching rules, fill with top-frequency non-matching rules
+
+Output format (JSON):
+```json
+{
+  "selectedRules": ["rule-name-1", "rule-name-2"],
+  "totalTokens": 2048,
+  "strategy": "frequency-first: selected 3 matching rules (2048 tokens), 1 backfill"
+}
+```
+
+Constraints:
+- Never exceed `contextBudget`
+- Always output valid JSON
+- Prefer fewer large rules over many small ones if they have equal frequency

package/genes/rule-router-relevance/.gene-manifest.json

@@ -0,0 +1,8 @@
+{
+  "geneId": "025849feb075fcb61916c89f0b7b58faeffbf2aed8d54cd83ef1a76a584c2391",
+  "name": "rule-router-relevance",
+  "domain": "prompt.rule-router",
+  "fidelity": "Wrapped",
+  "wrappedAt": "2026-03-24T01:54:54.000Z",
+  "fromSkill": null
+}

package/genes/rule-router-relevance/phenotype.json

@@ -0,0 +1,76 @@
+{
+  "domain": "prompt.rule-router",
+  "description": "Rule routing strategy that prioritizes semantic relevance. Scores each rule by triggerWord overlap density with the user message, injecting the most contextually relevant rules first. Variant B of the Rule Router Gene competition.",
+  "inputSchema": {
+    "type": "object",
+    "properties": {
+      "userMessage": {
+        "type": "string",
+        "description": "The user's current message/query"
+      },
+      "availableRules": {
+        "type": "array",
+        "items": {
+          "type": "object",
+          "properties": {
+            "name": { "type": "string" },
+            "triggerWords": { "type": "array", "items": { "type": "string" } },
+            "frequency": { "type": "integer", "description": "Historical trigger count" },
+            "sizeTokens": { "type": "integer" }
+          }
+        },
+        "description": "Available rules with metadata"
+      },
+      "contextBudget": {
+        "type": "integer",
+        "description": "Maximum tokens available for rule injection",
+        "default": 4096
+      }
+    },
+    "required": ["userMessage", "availableRules"]
+  },
+  "outputSchema": {
+    "type": "object",
+    "properties": {
+      "selectedRules": {
+        "type": "array",
+        "items": { "type": "string" },
+        "description": "Ordered list of rule names to inject"
+      },
+      "totalTokens": {
+        "type": "integer",
+        "description": "Total token cost of selected rules"
+      },
+      "strategy": {
+        "type": "string",
+        "description": "Routing strategy explanation"
+      }
+    }
+  },
+  "llmRequirements": {
+    "templateFormat": "mustache",
+    "templateVariables": ["userMessage", "availableRules", "contextBudget"],
+    "targetModels": ["claude-3", "gpt-4"],
+    "minContextWindow": 4096,
+    "expectedOutputFormat": "json",
+    "temperatureHint": 0.1,
+    "maxOutputTokens": 1024,
+    "systemPromptPath": "system-prompt.md",
+    "chainOfThought": true
+  },
+  "fitnessConfig": {
+    "evaluationMethod": "template-quality",
+    "dimensions": {
+      "routeHitRate": { "weight": 0.40, "description": "Fraction of injected rules actually triggered by user message" },
+      "contextSavings": { "weight": 0.30, "description": "1 - (usedTokens / contextBudget)" },
+      "templateValidity": { "weight": 0.20, "description": "Template renders without error for all valid inputs" },
+      "crossModelConsistency": { "weight": 0.10, "description": "Same input produces same rule selection across models" }
+    }
+  },
+  "dependencies": [],
+  "version": "0.1.0",
+  "author": "rotifer-team",
+  "fidelity": "Wrapped",
+  "transparency": "Open",
+  "source": "development-genome"
+}

package/genes/rule-router-relevance/system-prompt.md

@@ -0,0 +1,29 @@
+You are a Rule Router that selects which .cursor/rules to inject into the AI context.
+
+Your strategy: **RELEVANCE-FIRST ROUTING**
+
+Given a user message and a list of available rules, select rules by computing a relevance score for each rule based on triggerWord overlap with the user message.
+
+Algorithm:
+1. For each rule, compute `relevanceScore = matchedTriggerWords / totalTriggerWords`
+2. Apply a bonus: if multiple trigger words match, score += 0.1 per additional match
+3. Sort rules by `relevanceScore` (descending), breaking ties by `frequency`
+4. Select rules in order until `contextBudget` is exhausted
+5. Only include rules with `relevanceScore > 0` (no backfill with unrelated rules)
+
+Think step by step about which trigger words match the user's intent before scoring.
+
+Output format (JSON):
+```json
+{
+  "selectedRules": ["rule-name-1", "rule-name-2"],
+  "totalTokens": 1536,
+  "strategy": "relevance-first: 2 rules with scores [0.8, 0.5], total 1536 tokens"
+}
+```
+
+Constraints:
+- Never exceed `contextBudget`
+- Always output valid JSON
+- Prefer precision over recall — only inject rules that are demonstrably relevant
+- Zero-match rules are never selected (unlike frequency-first which backfills)