Export Pharma Packaging Compliance

SKILL.md - Skill Registry & Execution Guide

Skill Overview

export-pharma-packaging-compliance is a production-grade harness for Pharmaceutical Packaging Regulatory Compliance analysis. It transforms Claude into a domain expert that delivers structured, evidence-backed outputs through a systematic orchestration of specialized sub-skills.

Skill Registration

Skill Metadata

| Property | Value | |----------|-------| | Name | export-pharma-packaging-compliance | | Version | 1.0.0 | | Type | Domain Analysis Harness | | Domain | Pharmaceutical Packaging Regulatory Compliance | | Primary Trigger | /export-pharma-packaging-compliance | | Secondary Triggers | "pharma packaging compliance", "export packaging regulations", "DSCSA", "FMD", "ISO 15378" |

Entry Point

The skill is invoked via the main harness in skills/main.md. All execution flows through this orchestrator.

Architecture Pattern

Modular Chain-of-Thought Routing

The skill uses a sequential step-based routing pattern with explicit quality gates between each step:

User Input
    ↓
[Pre-Flight: Language Detection]
    ↓
Step 1: Requirements Gathering
    ↓ [Gate: Object Confirmed]
Step 2: Evidence Collection
    ↓ [Gate: Current Data + 1 Authoritative Doc]
Step 3: Core Analysis
    ↓ [Gate: Barrier/CR/Serialization/GMP]
Step 4: Knowledge Query
    ↓ [Gate: ≥1 Academic Source]
Step 5: Advisory Synthesis
    ↓ [Gate: Verdict Category + Disclosure]
Step 6: Quality Gate Review
    ↓ [Gates U1-U6, G1-G4]
Final Output

Graceful Degradation

The skill implements 5-level degradation:

• Level 0: Full analysis (all sources available)
• Level 1: Partial source failure (substitute with flag)
• Level 2: Most sources failed (knowledge base only)
• Level 3: Missing inputs (proceed with flags)
• Level 4: Complete failure (emit "DATA UNAVAILABLE")

Error Recovery

8 error types with explicit recovery strategies and retry limits.

Sub-Skill Registry

Sub-Skill Catalog

| ID | Name | Step | Timeout | Purpose | |----|------|------|---------|---------| | sub-gather-requirements | Requirements Gatherer | 1 | 60s | Clarify object, scope, constraints, language | | sub-evidence-collector | Evidence Collector | 2 | 120s | Fetch real-time and reference data | | sub-core-analysis | Core Analyzer | 3 | 180s | Verify compliance against standards | | sub-knowledge-updater | Knowledge Querier | 4 | 90s | Query academic knowledge base | | sub-advisor | Senior Advisor | 5 | 120s | Synthesize risk-disclosed conclusion |

Sub-Skill Resolution

Sub-skills are resolved by the main harness using the Skill tool with explicit skill path:

Skill("sub-gather-requirements")
Skill("sub-evidence-collector")
Skill("sub-core-analysis")
Skill("sub-knowledge-updater")
Skill("sub-advisor")

Sub-Skill Input/Output Contracts

Each sub-skill follows the JSON schema defined in config/schemas.json:

• RequirementsOutput: Object, scope, timeframe, inputs, audience, language
• EvidenceOutput: Current data, authoritative docs, recent news, benchmarks
• CoreAnalysisOutput: Barrier, CR/tamper, serialization, stability, GMP, scenarios
• AdvisorOutput: Verdict, scenarios, risks, evidence chain, remediation, disclosure

Tool Definitions

Tool Schemas

The skill uses the following tools with defined schemas:

WebSearch

• Purpose: Fetch real-time domain news and updates
• Timeout: 30 seconds
• Max Results: 10
• Schema:

{
  "query": "string",
  "maxResults": "integer (1-10)",
  "recency": "string (optional)"
}

WebFetch

• Purpose: Scrape authoritative domain sources
• Timeout: 30 seconds
• Max Retries: 3
• Schema:

{
  "url": "string (https required)",
  "format": "markdown (default)"
}

Read

• Purpose: Read SECOND-KNOWLEDGE-BRAIN.md
• Max File Size: 10MB
• Supported Formats: .md, .txt, .json

Write

• Purpose: Append knowledge entries
• Atomic: Yes
• Schema:

{
  "path": "string",
  "content": "string",
  "mode": "append (default)"
}

Bash

• Purpose: Execute knowledge_updater.py
• Timeout: 60 seconds
• Allowed Commands: python, pip, ls, cat, echo
• Schema:

{
  "command": "string",
  "timeout": "integer (default 60)"
}

Hooks System

Pre-Execution Hooks

Handlers executed before skill execution:

| Handler | Purpose | Implementation | |---------|---------|----------------| | languageDetection | Detect user language (en/vi) | Character analysis + domain word matching | | inputValidation | Validate required inputs present | Schema validation against RequirementsInput |

Post-Execution Hooks

Handlers executed after skill execution:

| Handler | Purpose | Implementation | |---------|---------|----------------| | qualityGateCheck | Verify all gates passed | Iterate U1-U6, G1-G4 with auto-fix | | outputFormatting | Ensure output format compliance | Validate against FinalReport schema |

Error Hooks

Handlers executed on error:

| Handler | Purpose | Implementation | |---------|---------|----------------| | errorLogging | Log error with context | Structured logging with metadata | | degradationEscalate | Escalate degradation level | Level-based behavior activation | | userNotification | Notify user of limitation | Emit limitation banner |

Quality Gates

Universal Gates (U1-U6)

| Gate | Check | Auto-Fix | Enforcement | |------|-------|----------|-------------| | U1 | ≥3 sources, ≥1 academic | Fetch from KB/evidence | Append before delivery | | U2 | Disclosure before recommendation | Prepend disclosure | Block until present | | U3 | Evidence hierarchy per source | Annotate tiers | Tag each source | | U4 | Language matches preference | Translate output | Re-detect language | | U5 | Output uses template | Reformat to template | Check sections | | U6 | Claims traceable to sources | Flag unsupported | Mark with [judgment] |

Domain Gates (G1-G4)

| Gate | Check | Auto-Fix | Enforcement | |------|-------|----------|-------------| | G1 | Barrier (MVTR/OTR) verified | Verify barrier | Add barrier check | | G2 | Child-resistant/tamper-evident | Add CR/tamper | Add CR/tamper check | | G3 | Serialization/traceability | Add serialization | Add serialization check | | G4 | GMP (ISO 15378) & validation | Add GMP/validation | Add GMP check |

Gate Enforcement Logic

for gate in gates:
    retry_count = 0
    while retry_count  [mandatory notice before recommendation]

## Recommendation / Conclusion
[verdict, scenarios, risks, evidence chain, remediation]

## Post-Execution Gate Checklist
[U1✓ U2✓ U3✓ U4✓ U5✓ U6✓ G1✓ G2✓ G3✓ G4✓ | Limitations: ...]

Language Support

Supported Languages

• English (en): Default
• Vietnamese (vi): Full translation support

Language Detection

Method: Character analysis + domain word matching

Vietnamese Indicators:

• Characters: à á ả ã ạ ă â đ è é ê ì í ò ó ô ơ ù ú ư ý
• Domain words: dược liệu, bao bì, tuân thủ, quy định, xuất khẩu

Fallback: Default to English, ask user to confirm

Translation Table

See config/config.json → i18n.translations for complete mapping.

Knowledge Base Integration

SECOND-KNOWLEDGE-BRAIN.md

Purpose: Living knowledge base updated by crawl pipeline

Structure:

1. Core Concepts & Frameworks
2. Key Research Papers & Standards
3. State-of-the-Art Methods
4. Authoritative Data Sources
5. Analytical Frameworks
6. Self-Update Protocol
7. Knowledge Update Log

Update Schedule:

• Academic: Weekly (Mondays 08:00)
• News: Daily (07:00)

Knowledge Query Pattern

def query_knowledge_base(keywords):
    entries = []
    for keyword in keywords:
        matches = search_brain(keyword)
        for match in matches:
            entries.append({
                "title": match.title,
                "authors": match.authors,
                "year": match.year,
                "venue": match.venue,
                "doi_or_url": match.doi_or_url,
                "tier": match.tier,
                "relevance_score": match.score
            })
    return sorted(entries, key=lambda x: x.relevance_score, reverse=True)[:5]

Monitoring & Logging

Metrics Collected

• executionTime: Time per step and total
• tokenUsage: Input/output tokens per step
• qualityGatePasses: Pass/fail per gate
• sourceAvailability: Success rate per source
• errorRates: Error type frequency

Log Format

{
  "timestamp": "2026-07-15T10:30:00Z",
  "level": "info",
  "component": "main",
  "event": "step_completed",
  "data": {
    "step": "sub-core-analysis",
    "duration_ms": 1523,
    "tokens_used": 8432,
    "gates_passed": ["G1", "G2", "G3", "G4"]
  }
}

Feature Flags

| Flag | Default | Purpose | |------|---------|---------| | enableKnowledgeUpdate | true | Enable crawl pipeline | | enableGracefulDegradation | true | Enable 5-level degradation | | enableAutoFix | true | Enable gate auto-fix | | enableMultiLanguage | true | Enable vi/en support | | enableCaching | true | Enable result caching | | enableParallelExecution | false | Enable parallel step execution |

Validation

Pre-Flight Validation

Before execution, validate:

1. Configuration files exist and are valid JSON
2. SECOND-KNOWLEDGE-BRAIN.md is accessible
3. Sub-skill files are present
4. Tools are available

Post-Execution Validation

After execution, validate:

1. Output matches FinalReport schema
2. All quality gates passed (or limitations flagged)
3. Language matches user preference
4. Disclosure is present before recommendation

Testing

Test Scenarios

See tests/test-scenarios.md for 5+ test scenarios:

• Standard compliance analysis
• Minimal input handling
• Comparison/conflict resolution
• Risk/conflict assessment
• Degraded mode operation

Test Execution

# Run all tests
python tools/run_test_scenarios.py --all

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

# Validate project structure
python tools/validate_project.py

Dependencies

Runtime Dependencies

requests>=2.31.0
feedparser>=6.0.10
python-dateutil>=2.8.2

Development Dependencies

pytest>=7.4.0
pytest-cov>=4.1.0
black>=23.7.0
flake8>=6.0.0
mypy>=1.5.0

Troubleshooting

Common Issues

Issue: "No CodeGraph project is loaded"

• Cause: Working directory detection issue
• Fix: Pass projectPath parameter explicitly

Issue: "Source timeout"

• Cause: Network connectivity or source unavailability
• Fix: Applies degradation level 1, retries with alternate source

Issue: "Knowledge base miss"

• Cause: No matches for query in SECOND-KNOWLEDGE-BRAIN.md
• Fix: Triggers gap-fill WebSearch, queues for crawl

Issue: "Quality gate failed after retries"

• Cause: Cannot auto-fix the gate issue
• Fix: Emits limitation notice, continues with flag

License

MIT License - See LICENSE file

Version History

• v1.0.0 (2026-07-15): Initial production release

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/export-pharma-packaging-compliance-agent-skill
• License: MIT

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