periplon 0.2.0

# Stdio and Context Management Example Workflow
#
# This workflow demonstrates the new stdio and context management features:
# - Output capture with truncation
# - Smart context injection based on dependencies
# - Configurable cleanup strategies
# - Task-level overrides

name: "Stdio Context Management Demo"
version: "1.0.0"
dsl_version: "1.0.0"

# Workflow-level limits apply to all tasks unless overridden
limits:
  # Output capture limits
  max_stdout_bytes: 524288        # 512KB - prevents unbounded output
  max_stderr_bytes: 131072        # 128KB
  max_combined_bytes: 655360      # 640KB total
  truncation_strategy: tail       # Keep last N bytes (most recent output)

  # Context injection limits
  max_context_bytes: 102400       # 100KB - limits context size
  max_context_tasks: 10           # Maximum 10 tasks in context

  # External storage (for very large outputs)
  external_storage_threshold: 5242880  # 5MB - store larger outputs externally
  external_storage_dir: ".workflow_outputs"
  compress_external: true

  # Cleanup strategy - keep most relevant outputs
  cleanup_strategy:
    type: highest_relevance
    keep_count: 15

# Define agents with specific capabilities
agents:
  data_fetcher:
    description: "Fetches large datasets from external sources"
    model: "claude-sonnet-4-5"
    tools: [WebSearch, WebFetch, Read, Write]

  data_processor:
    description: "Processes and transforms data"
    model: "claude-sonnet-4-5"
    tools: [Read, Write, Bash]

  analyzer:
    description: "Analyzes processed data and generates insights"
    model: "claude-sonnet-4-5"
    tools: [Read, Write]

  reporter:
    description: "Creates final reports"
    model: "claude-sonnet-4-5"
    tools: [Read, Write]

tasks:
  # Task 1: Fetch data (may produce large output)
  fetch_data:
    description: "Fetch dataset from external API"
    agent: "data_fetcher"
    output: "raw_data.json"

    # Override workflow limits for this verbose task
    limits:
      max_stdout_bytes: 2097152    # 2MB - this task produces more output
      truncation_strategy: both     # Keep first and last portions

    script:
      language: bash
      content: |
        # Simulate fetching large dataset
        echo "Fetching data from API..."
        curl -s https://api.example.com/large-dataset > raw_data.json
        echo "Downloaded $(wc -l < raw_data.json) records"

  # Task 2: Clean and validate data
  clean_data:
    description: "Clean and validate the fetched data"
    agent: "data_processor"
    depends_on: [fetch_data]
    output: "clean_data.json"

    # Automatic context mode - only includes 'fetch_data' output (direct dependency)
    context:
      mode: automatic
      min_relevance: 0.8    # Only very relevant context (direct dependencies)
      max_bytes: 50000      # Limit context to 50KB

    script:
      language: bash
      content: |
        # Clean and validate data
        python3 scripts/clean_data.py raw_data.json clean_data.json
        echo "Cleaned data: $(jq length clean_data.json) valid records"

  # Task 3: Transform data
  transform_data:
    description: "Transform data into analysis-ready format"
    agent: "data_processor"
    depends_on: [clean_data]
    output: "transformed_data.json"

    # Manual context control - explicitly include only what we need
    context:
      mode: manual
      include_tasks: [clean_data]    # Only include clean_data output
      exclude_tasks: [fetch_data]    # Explicitly exclude raw fetch data
      max_bytes: 30000

    script:
      language: bash
      content: |
        python3 scripts/transform.py clean_data.json transformed_data.json
        echo "Transformed $(jq length transformed_data.json) records"

  # Task 4: Statistical analysis
  analyze_stats:
    description: "Perform statistical analysis on the transformed data"
    agent: "analyzer"
    depends_on: [transform_data]
    output: "statistics.json"

    # Default automatic context - gets transform_data (relevance=1.0)
    # and clean_data (relevance=0.8/depth=0.4) due to transitive dependency

    script:
      language: bash
      content: |
        python3 scripts/analyze.py transformed_data.json statistics.json
        cat statistics.json

  # Task 5: Pattern detection (parallel with stats)
  detect_patterns:
    description: "Detect patterns and anomalies in the data"
    agent: "analyzer"
    depends_on: [transform_data]
    parallel_with: [analyze_stats]
    output: "patterns.json"

    script:
      language: bash
      content: |
        python3 scripts/detect_patterns.py transformed_data.json patterns.json
        echo "Found $(jq '.patterns | length' patterns.json) patterns"

  # Task 6: Generate insights (depends on both analysis tasks)
  generate_insights:
    description: "Generate insights from statistical analysis and patterns"
    agent: "analyzer"
    depends_on: [analyze_stats, detect_patterns]
    output: "insights.md"

    # Automatic context includes both direct dependencies
    context:
      mode: automatic
      min_relevance: 0.5
      max_tasks: 5

  # Task 7: Create final report
  create_report:
    description: "Create comprehensive final report"
    agent: "reporter"
    depends_on: [generate_insights]
    output: "final_report.md"

    # No context needed - only uses the insights file
    context:
      mode: none

    limits:
      max_stdout_bytes: 1048576    # 1MB for report generation
      truncation_strategy: summary  # Use AI summarization if output is large

  # Task 8: Verify output (runs after report)
  verify_output:
    description: "Verify all outputs were generated correctly"
    agent: "data_processor"
    depends_on: [create_report]

    script:
      language: bash
      content: |
        # Verify all expected files exist
        for file in raw_data.json clean_data.json transformed_data.json statistics.json patterns.json insights.md final_report.md; do
          if [ -f "$file" ]; then
            echo "✓ $file exists ($(du -h $file | cut -f1))"
          else
            echo "✗ $file missing"
            exit 1
          fi
        done
        echo "All outputs verified successfully"

# Workflow orchestration
workflows:
  main:
    description: "Main data processing workflow"
    tasks: [fetch_data, clean_data, transform_data, analyze_stats, detect_patterns, generate_insights, create_report, verify_output]
    execution_mode: sequential