curator

VERSION: 3.1.0

name: curator description: "Full curator pipeline for autonomous learning from quality repositories. Executes: discovery → scoring → ranking → ingest → learn → vault sync. Writes to procedural memory AND Obsidian vault for Graph View visualization and graduation pipeline. Use for: populating procedural memory with domain patterns, first-time domain learning, comprehensive knowledge building. Triggers: /curator full, 'learn patterns from repos', 'build knowledge base'." argument-hint: "[full|quick|status] --type --lang " user-invocable: true context: fork allowed-tools: - Task - Bash - Read - Write - WebSearch - WebFetch

Curator Pipeline Skill (v3.1.0)

Full Autonomous Learning Pipeline - Discovers, scores, and learns from quality repositories.

Role & Priorities

Priorities (ordered): quality → coverage → relevance → performance → speed

Scope: Repository discovery, quality scoring, pattern extraction, procedural memory population.

Agent Teams Integration (v2.88)

Optimal Scenario: C (Integrated)

Why Scenario C for Curator

• High coordination need: 5+ sequential pipeline stages
• Quality gates required: Each stage needs validation before proceeding
• Multi-tool operations: GitHub API, git, file processing, JSON manipulation
• Scalability: Can process multiple repositories in parallel

Scenario Analysis

Workflow (Scenario C)

# Integrated Team Workflow
TeamCreate(team_name="curator-pipeline", description="Learning from ${DOMAIN} repos")

# Stage 1: Discovery
Task(subagent_type="ralph-researcher", prompt="Search GitHub for ${DOMAIN} repositories")
→ Returns candidate list

# Stage 2: Scoring (parallel)
Task(subagent_type="ralph-reviewer", prompt="Score ${REPO_1} quality")
Task(subagent_type="ralph-reviewer", prompt="Score ${REPO_2} quality")
→ Returns quality scores

# Stage 3: Ranking
Team lead aggregates scores and selects top N

# Stage 4: Ingest & Learn (parallel)
Task(subagent_type="ralph-coder", prompt="Clone and extract patterns from ${TOP_REPO}")
→ Returns extracted patterns

# Stage 5: Quality Gate
TeammateIdle hook validates pattern quality
TaskCompleted hook verifies manifest population

# Stage 6: Injection
Procedural memory updated automatically

Pipeline Stages

1. Discovery (curator-discovery.sh)

# Search GitHub for repositories
--type <domain>    # backend, frontend, database, security, devops, testing
--lang <language>  # typescript, python, go, rust, java
--tier <tier>      # premium (1000+ stars), standard (500+), economic (100+)

Output: Candidate repository list with metadata.

2. Scoring (curator-scoring.sh)

Quality metrics:

• Star count and trend
• Recent commit activity
• Documentation quality
• Test coverage indicators
• Organization reputation

Output: Scored repository list (0-100).

3. Ranking (curator-rank.sh)

# Select top repositories
--max <n>         # Maximum repos to process (default: 3)
--diversity       # Ensure organization diversity

Output: Ranked candidate list.

4. Ingest (curator-ingest.sh)

# Clone and prepare repositories
--clone-depth 1   # Shallow clone for efficiency

Output: Cloned repositories in corpus/pending/.

5. Approve (curator-approve.sh)

# Manual or automatic approval
--auto            # Auto-approve based on score threshold
--threshold 75    # Minimum score for auto-approval

Output: Repositories moved to corpus/approved/.

6. Learn (curator-learn.sh) - GAP FIXES v2.88

# Extract patterns and populate procedural memory
# GAP-C01 FIX: Manifest files[] now populated
# GAP-C02 FIX: Domain detection and assignment

Output:

• Updated .claude/rules/learned/ (MemPalace taxonomy)
• Manifest with files[] array
• Domain-categorized rules

Commands

Full Pipeline

/curator full --type backend --lang typescript

Executes all stages: discovery → scoring → ranking → ingest → approve → learn.

Quick Pipeline

/curator quick --type security --lang python --repo owner/repo

Skips discovery, learns from specific repository.

Status Check

/curator status

Shows:

• Approved repositories count
• Rules per domain
• Learning gaps

Configuration

// ~/.ralph/config/memory-config.json
{
  "curator": {
    "max_repos_per_run": 3,
    "min_stars": 100,
    "clone_depth": 1,
    "auto_approve_threshold": 75,
    "domains": ["backend", "frontend", "database", "security", "devops", "testing"]
  },
  "auto_learn": {
    "enabled": true,
    "blocking": false,
    "min_rules_domain": 3
  }
}

Quality Gates (v2.88)

GAP Fixes Applied (v2.88)

GAP-C01: Manifest Files[] Population

Before:

{"files": [], "patterns_extracted": 0}

After:

{
  "files": ["src/handler.ts", "src/middleware.ts"],
  "patterns_extracted": 5,
  "detected_domain": "backend",
  "detected_language": "typescript"
}

GAP-C02: Domain Detection

Rules now automatically categorized:

• Keyword analysis of repository content
• File extension detection
• Configuration file inspection

Related Skills

• /curator-repo-learn - Single repository learning (Scenario B)
• /repo-learn - Alias for curator-repo-learn
• /smart-fork - Pattern extraction from external repos

Hooks Integration

Action Reporting (v2.93.0)

Esta skill genera reportes automáticos completos para trazabilidad:

Reporte Automático

Cuando esta skill completa, se genera automáticamente:

1. En la conversación de Claude: Resultados visibles
2. En el repositorio: docs/actions/curator/{timestamp}.md
3. Metadatos JSON: .claude/metadata/actions/curator/{timestamp}.json

Contenido del Reporte

Cada reporte incluye:

• ✅ Summary: Descripción de la tarea ejecutada
• ✅ Execution Details: Duración, iteraciones, archivos modificados
• ✅ Results: Errores encontrados, recomendaciones
• ✅ Next Steps: Próximas acciones sugeridas

Ver Reportes Anteriores

# Listar todos los reportes de esta skill
ls -lt docs/actions/curator/

# Ver el reporte más reciente
cat $(ls -t docs/actions/curator/*.md | head -1)

# Buscar reportes fallidos
grep -l "Status: FAILED" docs/actions/curator/*.md

Generación Manual (Opcional)

source .claude/lib/action-report-lib.sh
start_action_report "curator" "Task description"
# ... ejecución ...
complete_action_report "success" "Summary" "Recommendations"

Referencias del Sistema

• Action Reports System - Documentación completa

• action-report-lib.sh - Librería helper

• action-report-generator.sh - Generador

• Learning System Audit

• Learning System Scenarios

• MULTI_AGENT_SCENARIOS_v2.88