Mangrove Forest Ai Conservation

Mangrove Forest AI Conservation Skill

Skill Registry & Metadata

| Property | Value | |----------|-------| | Skill ID | mangrove-forest-ai-conservation | | Version | 1.0.0 | | Status | Production Ready | | Primary Domain | Mangrove Forest Remote Sensing & Conservation | | Dependencies | Python 3.11+, Claude Code with skill support | | Compatibility | Claude Code, Claude.ai (with adaptations) |

Skill Resolution & Execution

Resolution Path

When a user invokes /mangrove-forest-ai-conservation, Claude Code resolves this skill through:

1. Name Resolution: Matches skill name in available skills registry
2. File Location: Loads skills/main.md as entry point
3. Dependency Check: Verifies SECOND-KNOWLEDGE-BRAIN.md availability
4. Language Detection: Pre-flight Vietnamese/English detection
5. Sub-Skill Chain: Sequentially invokes required sub-skills

Input Schema

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "title": "MangroveConservationInput",
  "type": "object",
  "properties": {
    "query": {
      "type": "string",
      "description": "User's analysis request",
      "examples": [
        "Analyze mangrove forest extent and health in the Mekong Delta",
        "Assess deforestation rates in the Sundarbans from 2015 to present"
      ]
    },
    "language": {
      "type": "string",
      "enum": ["en", "vi", "auto"],
      "default": "auto",
      "description": "Output language (English/Vietnamese/auto-detect)"
    },
    "region": {
      "type": "string",
      "description": "Geographic region for analysis (optional if specified in query)"
    },
    "timeframe": {
      "type": "string",
      "description": "Analysis timeframe (optional)"
    },
    "analysis_type": {
      "type": "string",
      "enum": ["standard", "deforestation", "restoration", "carbon", "health", "comprehensive"],
      "default": "comprehensive"
    }
  },
  "required": ["query"]
}

Output Schema

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "title": "MangroveConservationReport",
  "type": "object",
  "properties": {
    "report_type": {
      "type": "string",
      "const": "AI-Based Mangrove Forest Monitoring & Conservation — Report"
    },
    "metadata": {
      "type": "object",
      "properties": {
        "date": {"type": "string", "format": "date"},
        "analyst": {"type": "string", "pattern": "^mangrove-forest-ai-conservation v"},
        "language": {"type": "string", "enum": ["English", "Tiếng Việt"]}
      }
    },
    "executive_summary": {"type": "string"},
    "inputs_scope": {
      "type": "object",
      "properties": {
        "region": {"type": "string"},
        "timeframe": {"type": "string"},
        "analysis_type": {"type": "string"}
      }
    },
    "evidence_collected": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "source": {"type": "string"},
          "tier": {"type": "string", "enum": ["Tier 1", "Tier 2", "Tier 3", "Tier 4"]},
          "date": {"type": "string"},
          "findings": {"type": "string"}
        }
      }
    },
    "analysis_scorecard": {
      "type": "object",
      "properties": {
        "extent_change": {"type": "string"},
        "deforestation_hotspots": {"type": "string"},
        "health_status": {"type": "string"},
        "species_composition": {"type": "string"},
        "blue_carbon": {"type": "string"},
        "restoration_suitability": {"type": "string"}
      }
    },
    "academic_evidence": {
      "type": "array",
      "items": {
        "type": "object",
        "properties": {
          "citation": {"type": "string"},
          "tier": {"type": "string"},
          "relevance": {"type": "string"}
        }
      }
    },
    "disclosure_limitations": {"type": "string"},
    "verdict": {
      "type": "string",
      "enum": [
        "Strong Conservation Plan",
        "Conditional (data gaps)",
        "High Degradation",
        "Inconclusive"
      ]
    },
    "scenarios": {
      "type": "object",
      "properties": {
        "best": {"type": "string"},
        "base": {"type": "string"},
        "worst": {"type": "string"}
      }
    },
    "quality_gates_passed": {
      "type": "array",
      "items": {"type": "string", "pattern": "^[UuG][1-6]$"}
    }
  },
  "required": ["report_type", "metadata", "executive_summary", "verdict", "disclosure_limitations"]
}

Harness Architecture

┌─────────────────────────────────────────────────────────────────┐
│                     USER INPUT                                   │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 0] Pre-Flight: Language Detection & Input Validation      │
│  Detect Vietnamese/English → Validate input schema               │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 1] sub-gather-requirements                               │
│  Clarify object, scope, timeframe, inputs, audience, language   │
│  Quality Gate: ≥1 object confirmed                              │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 2] sub-evidence-collector                                │
│  Fetch authoritative real-time and reference data                │
│  Quality Gate: current data + 1 doc OR limitation flag          │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 3] sub-core-analysis                                     │
│  Execute extent mapping, deforestation detection, health          │
│  assessment, carbon estimation, restoration planning             │
│  Quality Gate: extent + deforestation + health + carbon           │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 4] sub-knowledge-updater                                 │
│  Query SECOND-KNOWLEDGE-BRAIN.md for academic evidence           │
│  Quality Gate: ≥1 academic source + coverage rating              │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 5] sub-advisor                                            │
│  Synthesize verdict, scenarios, risks, evidence chain            │
│  Quality Gate: verdict in set + disclosure before conclusion     │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│  [STEP 6] Quality Gate Review (Main Harness)                     │
│  Verify U1-U6 + G1-G4 with auto-fix & enforcement               │
│  Exit: All gates pass OR explicit limitation flagged             │
└────────────────────────┬────────────────────────────────────────┘
                         │
                         ▼
┌─────────────────────────────────────────────────────────────────┐
│                   FINAL OUTPUT DELIVERY                          │
│  Full report with evidence citations, disclosure, scenarios      │
└─────────────────────────────────────────────────────────────────┘

Validation & Quality Assurance

Quality Gates

Universal Gates (U1-U6):

• U1: ≥3 sources cited, ≥1 academic/authoritative
• U2: Disclosure/limitations before recommendation
• U3: Evidence hierarchy stated per source (Tier 1-4)
• U4: Language matches user preference
• U5: Output uses declared template (all sections)
• U6: Every claim traceable to ≥1 source or flagged

Domain Gates (G1-G4):

• G1: Extent mapped with AI/satellite
• G2: Deforestation/degradation detected
• G3: Health & blue carbon assessed
• G4: Restoration suitability & threats identified

Auto-Fix Logic

Each gate has:

• Auto-Fix Procedure: Action to remediate failed gate
• Max Retries: 2 attempts per gate
• Fallback Strategy: Explicit limitation flag on failure
• Enforcement Level: Block / Warn / Continue

Output Validation

The skill validates output against:

1. Structural Validation: All required sections present
2. Content Validation: Minimum evidence thresholds met
3. Schema Validation: Output matches declared JSON schema
4. Language Validation: Output language matches detection
5. Disclosure Validation: Limitations appear before recommendations

Error Handling & Graceful Degradation

Degradation Levels

| Level | Description | Limitation Banner | |-------|-------------|-------------------| | 0 | Full data available | None | | 1 | Some secondary sources unavailable | Minor limitation notice | | 2 | Primary sources degraded | Moderate limitation notice | | 3 | Most sources unavailable | Severe limitation notice | | 4 | All sources failed | Critical limitation + DATA UNAVAILABLE |

Error Recovery

Each sub-skill implements:

1. Primary Source Attempt: Try authoritative sources first
2. Fallback Chain: Secondary → Knowledge base → Cached data
3. Explicit Flagging: Clear DATA UNAVAILABLE markers
4. Never Silent: Always indicate what's missing

Integration Points

Required Files

• skills/main.md — Main harness orchestrator
• skills/sub-*.md — Domain sub-skills (5 files)
• SECOND-KNOWLEDGE-BRAIN.md — Living knowledge base
• tools/knowledge_updater.py — Crawl pipeline

Optional Integration

• config/settings.yaml — Configuration management
• references/*.md — Domain knowledge templates
• scripts/*.py — Automation utilities
• hooks/*.py — Lifecycle event handlers

External APIs

• Global Mangrove Watch (web scraping)
• Semantic Scholar (academic search)
• ArXiv (preprint server)
• Domain RSS feeds (news updates)

Configuration Management

Configuration is loaded from multiple sources in priority order:

1. Environment Variables: Override all other settings
2. config/settings.yaml: Project-specific configuration
3. config/defaults.yaml: Default fallback settings
4. Hardcoded defaults in KNOWLEDGE_CONFIG

Configuration Schema

# config/settings.yaml
language:
  default: en
  fallback: en

analysis:
  default_type: comprehensive
  max_sources: 10
  min_academic_sources: 1

knowledge_update:
  enabled: true
  weekly_schedule: "0 8 * * 1"
  daily_schedule: "0 7 * * *"

data_sources:
  timeout_seconds: 30
  retry_attempts: 3
  fallback_to_cache: true

quality_gates:
  strict_mode: false
  auto_fix_enabled: true
  max_retries: 2

Lifecycle Hooks

Available Hook Points

| Hook Point | Description | Use Case | |-----------|-------------|----------| | pre_analysis | Before step 1 | Input validation, logging | | post_requirements | After step 1 | Requirements confirmation | | post_evidence | After step 2 | Data quality check | | post_analysis | After step 3 | Intermediate results | | post_knowledge | After step 4 | Citation validation | | pre_delivery | Before step 6 | Final quality checks | | post_delivery | After output | Notification, metrics |

Hook Implementation

Hooks are Python modules in hooks/ directory with execute(context) function:

# hooks/example_hook.py
def execute(context: dict) -> dict:
    """Hook execution function
    
    Args:
        context: Analysis context with step, data, config
    
    Returns:
        Modified context or None for no changes
    """
    # Hook logic here
    return context

Performance Optimization

Token Management

• Progressive Disclosure: Load reference files on-demand
• Caching Strategy: Cache external API responses for 24h
• Batch Operations: Aggregate multiple API calls when possible
• Compression: Use compact JSON for intermediate data

Execution Optimization

• Parallel Sub-Skills: Independent sub-skills run in parallel where possible
• Early Exit: Skip unnecessary analysis based on requirements
• Lazy Loading: Load heavy resources only when needed
• Result Caching: Cache analysis results for similar queries

Testing & Validation

Test Coverage

• Unit Tests: Individual sub-skill validation
• Integration Tests: Full harness execution
• Scenario Tests: 5+ comprehensive scenarios
• Edge Cases: Degraded mode, minimal input, conflicts

Test Execution

# Run all tests
python tools/run_test_scenarios.py

# Run specific test
python tools/run_test_scenarios.py --scenario standard

# Dry run test
python tools/run_test_scenarios.py --dry-run

Contributing & Extension

Adding New Analysis Methods

1. Define method in sub-core-analysis.md
2. Add quality gate in main.md
3. Create test scenario in tests/test-scenarios.md
4. Update SECOND-KNOWLEDGE-BRAIN.md with references

Adding New Data Sources

1. Update KNOWLEDGE_CONFIG in tools/knowledge_updater.py
2. Add fallback logic in sub-evidence-collector.md
3. Update data sources table in README.md
4. Test with degraded source scenarios

Version History

| Version | Date | Changes | |---------|------|---------| | 1.0.0 | 2026-07-27 | Initial production release |

License

MIT License — See LICENSE file for details

Contact & Support

• Issues: GitHub Issues
• Documentation: README.md
• Technical Spec: PROJECT-detail.md

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: dungnotnull
• Source: dungnotnull/mangrove-forest-ai-conservation-agent-skill
• License: MIT

Install and usage instructions live in the source repository linked above.