Ancient Costume Mural Reconstruction

SKILL.md — Skill Registry & Architecture

What This Skill Does

This skill transforms Claude into a Senior Ancient Costume Reconstruction & Archaeological Textile History Specialist. When users ask about reconstructing garments from murals/statues, identifying historical textiles, analyzing period construction, recovering ancient dye recipes, or producing 3D reconstructions of archaeological dress, this skill:

1. Clarifies requirements (object, scope, timeframe, inputs, audience, language)
2. Fetches authoritative evidence (museum records, extant textiles, academic sources)
3. Applies domain methods (iconography → construction → materials → 3D reconstruction)
4. Surfaces academic evidence (Tier-labelled citations from SECOND-KNOWLEDGE-BRAIN)
5. Delivers risk-disclosed conclusions (evidence-graded verdict with full disclosure)

Key capability: This is NOT just costume design — it's archaeologically-grounded reconstruction that respects evidence hierarchies (extant textiles > iconography > textual > ethnographic) and always discloses limitations before conclusions.

---

When This Skill Triggers

This skill should trigger for ANY user query involving:

Direct Reconstruction Tasks

• "Reconstruct the [garment] from [mural/cave/tomb]"
• "What did [historical figure] wear in [painting/relief]?"
• "Build a 3D model of [period] costume based on [statue]"

Analysis & Identification

• "Analyze this textile fragment"
• "Identify the dye used in [archaeological find]"
• "What materials would this [period] garment be made of?"
• "Examine the construction of [garment in artwork]"

Comparative & Historical

• "Compare Han and Tang dynasty court robes"
• "How did [garment type] change from [period A] to [period B]?"
• "What's the evidence for [costume element] in [culture]?"

Museum & Collection Queries

• "Find museum records for [accession number]"
• "What extant [garment type] examples survive from [period]?"
• "Show me comparable textiles to [artifact]"

Materials & Techniques

• "What dyes were available in [period/region]?"
• "How was [fabric type] made in [culture]?"
• "Reconstruct the weave structure of [textile]"

Even if the user doesn't say "reconstruct" — if they're asking about historical garments, archaeological textiles, period dress, or museum costume collections, this skill should trigger.

---

Skill Registry Architecture

Registration Model

Skills are registered in assets/skill_manifest.json with:

{
  "version": "2.0.0",
  "skills": [
    {
      "name": "skill-name",
      "kind": "orchestrator|router|sub-agent",
      "path": "skills/skill-name.md",
      "description": "One-line summary",
      "step": N,
      "inputs_schema": "assets/schemas/xxx.schema.json",
      "outputs_schema": "assets/schemas/yyy.schema.json",
      "tools": ["Tool1", "Tool2"],
      "quality_gates": ["U1", "G1"],
      "routing": {...},
      "fallback_skill": "fallback-name"
    }
  ]
}

Skill Kinds

| Kind | Purpose | Examples | |------|---------|----------| | orchestrator | Top-level harness, manages workflow, quality gates | main.md | | router | Chain-of-thought routing to specialized agents | sub-router.md | | sub-agent | Domain-specialized analysis agent | All sub-*.md files |

Skill Resolution Pipeline

User invokes /ancient-costume-mural-reconstruction
    ↓
Load SKILL.md (this file) → skill_manifest.json
    ↓
Resolve execution chain via sub-router
    ↓
Sequential invocation:
  1. sub-gather-requirements (intake)
  2. sub-evidence-collector (data)
  3. sub-router (routing decision)
  4. [Specialized agents per routing plan]
  5. sub-knowledge-updater (academic evidence)
  6. sub-advisor (synthesis → verdict)
    ↓
Quality gate evaluation (U1-U6, G1-G4)
    ↓
Graceful degradation if gates fail
    ↓
Render final report via render_report tool
    ↓
Deliver to user

---

Input/Output JSON Schemas

Requirements Schema (inputs to Steps 1-2)

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "title": "RequirementsBundle",
  "type": "object",
  "required": ["object", "period", "scope"],
  "properties": {
    "object": {
      "type": "string",
      "description": "The mural/statue/relief/site or garment being analyzed"
    },
    "period": {
      "type": "string",
      "description": "Historical period/culture (e.g., 'Tang Dynasty', 'Han China')"
    },
    "scope": {
      "type": "string",
      "enum": ["iconography", "construction", "materials", "3d", "comparison", "combined"],
      "description": "Analysis type focus"
    },
    "timeframe": {
      "type": "string",
      "description": "Urgency or deadline constraints"
    },
    "available_inputs": {
      "type": "array",
      "items": {"type": "string"},
      "description": "Available sources (images, documents, artifacts)"
    },
    "target_audience": {
      "type": "string",
      "description": "Who will use this analysis"
    },
    "language": {
      "type": "string",
      "enum": ["en", "vi"],
      "description": "Output language"
    },
    "analysis_type": {
      "type": "string",
      "description": "Specific analysis type if different from scope"
    }
  }
}

Evidence Bundle Schema (outputs from Step 2)

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "title": "EvidenceBundle",
  "type": "object",
  "properties": {
    "current_collection_records": {
      "type": "array",
      "items": {"$ref": "https://example.org/evidence_item.schema.json"}
    },
    "extant_parallels": {
      "type": "array",
      "items": {"$ref": "https://example.org/evidence_item.schema.json"}
    },
    "authoritative_docs": {
      "type": "array",
      "items": {"$ref": "https://example.org/evidence_item.schema.json"}
    },
    "recent_developments": {
      "type": "array",
      "items": {"$ref": "https://example.org/evidence_item.schema.json"}
    },
    "reference_benchmarks": {
      "type": "object",
      "description": "Period-specific benchmarks from SECOND-KNOWLEDGE-BRAIN"
    }
  }
}

Reconstruction Schema (outputs from Step 4)

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "title": "ReconstructionReport",
  "type": "object",
  "required": ["iconographic_analysis", "construction_recovery", "materials_analysis", "evidence_hierarchy_applied", "3d_reconstruction"],
  "properties": {
    "iconographic_analysis": {
      "type": "object",
      "description": "Garment form, drape, layering, accessories, status markers"
    },
    "construction_recovery": {
      "type": "object",
      "description": "Pattern, seaming, draping, period techniques"
    },
    "materials_analysis": {
      "type": "object",
      "description": "Fiber identification, dye analysis, textile technology"
    },
    "evidence_hierarchy_applied": {
      "type": "boolean",
      "description": "Whether evidence hierarchy was explicitly applied"
    },
    "3d_reconstruction": {
      "type": "object",
      "description": "Layered 3D model with confidence levels and scenarios"
    },
    "cultural_context": {
      "type": "object",
      "description": "Social/cultural significance of the garment"
    }
  }
}

Verdict Schema (final output)

{
  "$schema": "http://json-schema.org/draft-07/schema#",
  "title": "ReconstructionVerdict",
  "type": "object",
  "required": ["verdict", "evidence_chain", "key_risks", "disclosure"],
  "properties": {
    "verdict": {
      "type": "string",
      "enum": [
        "Evidence-Based Reconstruction",
        "Plausible (interpretive)",
        "Speculative",
        "Inconclusive"
      ]
    },
    "scenarios": {
      "type": "array",
      "description": "Best/base/worst case scenarios"
    },
    "key_risks": {
      "type": "array",
      "minItems": 3,
      "items": {
        "type": "object",
        "properties": {
          "description": {"type": "string"},
          "probability": {"type": "string"},
          "impact": {"type": "string"}
        }
      }
    },
    "evidence_chain": {
      "type": "object",
      "description": "Full evidence chain from sources to conclusion"
    },
    "recommended_actions": {
      "type": "array",
      "items": {"type": "string"}
    },
    "disclosure": {
      "type": "string",
      "description": "MUST appear before the verdict in output"
    }
  }
}

---

Tool Definitions & Execution Handlers

Tool Registry

All tools are defined in assets/tool_definitions.json with:

• name: Tool identifier
• description: What the tool does
• category: "knowledge" | "retrieval" | "analysis" | "validation" | "io" | "utility"
• input_schema: Path to JSON schema for inputs
• output_schema: Path to JSON schema for outputs
• handler: Python function path in tools/agent_tools.py
• timeout_seconds: Maximum execution time
• idempotent: Whether repeated calls produce same result
• requires_network: Whether tool needs internet access

Tool Execution Flow

Agent invokes tool by name
    ↓
Load tool definition from tool_definitions.json
    ↓
Validate inputs against input_schema
    ↓
Call handler function in tools/agent_tools.py
    ↓
Enforce timeout (raises TimeoutError if exceeded)
    ↓
Validate outputs against output_schema
    ↓
Return structured result

Available Tools

| Tool | Handler | Category | Description | |------|---------|----------|-------------| | search_knowledge_base | search_knowledge_base() | knowledge | Search SECOND-KNOWLEDGE-BRAIN for Tier-labelled citations | | fetch_museum_record | fetch_museum_record() | retrieval | Fetch museum collection record by accession number | | build_3d_layer | build_3d_layer() | analysis | Build 3D reconstruction layer with material assignment | | queue_crawl_gap | queue_crawl_gap() | knowledge | Queue crawl gap for knowledge pipeline | | run_quality_gate | run_quality_gate() | validation | Evaluate quality gate (U1-U6, G1-G4) | | validate_verdict | validate_verdict() | validation | Validate verdict against schema | | render_report | render_report() | io | Render final report from components | | emit_event | emit_event() | utility | Emit lifecycle event to hooks bus |

---

Quality Gates (Validation Rules)

Universal Gates (U1-U6)

Apply to ALL harness outputs:

| Gate | Check | Auto-Fix | |------|-------|----------| | U1 | ≥3 sources cited, ≥1 Tier 1 (academic/authoritative) | Add 2+ sources, include Tier 1 | | U2 | Disclosure BEFORE recommendation | Move disclosure to appear first | | U3 | Evidence hierarchy stated per source | Add Tier labels (Tier 1-4) | | U4 | Language matches user preference | Ensure all text in declared language | | U5 | Output follows template | Add missing sections | | U6 | All claims traceable to source OR flagged as judgment | Add citations or flag as judgment |

Domain Gates (G1-G4)

Apply to ancient costume reconstruction specifically:

| Gate | Check | Auto-Fix | |------|-------|----------| | G1 | Iconographic analysis completed | Perform visual source analysis | | G2 | Construction & materials/dyes recovered | Add period construction & materials | | G3 | Evidence hierarchy applied per claim | State hierarchy (extant > iconographic > textual > ethnographic) | | G4 | 3D reconstruction produced with confidence | Build layered 3D model with H/M/L confidence |

Gate Enforcement Logic

For each gate:
  1. Run gate check against payload
  2. If PASS → continue
  3. If FAIL → apply auto-fix (if enabled) → retry
  4. If still FAIL after 2 retries → flag limitation → continue
  5. If enforcing mode (default) → fail the step → escalate degradation level

---

Graceful Degradation Strategy

Degradation Levels

| Level | Condition | Behavior | |-------|-----------|----------| | 0 | All gates pass | Full execution, no limitations | | 1 | Non-critical gates fail | Add limitation banner, continue | | 2 | Critical gates fail, fallback available | Use fallback agent, add limitation | | 3 | Major limitations | Best-effort analysis, heavy disclosure | | 4 | Severe failures | Minimum viable output, Inconclusive verdict |

Degradation Banners

When degradation occurs, output MUST include:

---
**LIMITATION**: This analysis has [data/method] limitations.
[Specific limitation description]
Confidence: [Low/Moderate/High]
Verdict may be affected: [Yes/No]
---

---

Hooks System (Lifecycle Events)

Event Types

Events are emitted via emit_event tool:

| Event Type | When | Payload | |------------|------|---------| | routing_decision | After sub-router completes | executionchain, rationale, fallback | | quality_gate_pass | After each gate passes | gateid, payload | | quality_gate_fail | After gate fails (with auto-fix) | gateid, failurereason, autofix | | degradation | When degradation level increases | oldlevel, newlevel, reason | | step_complete | After each sub-skill completes | skillname, outputs, durationms | | analysis_complete | After full harness completes | verdict, totalduration_ms |

Hook Handlers

Hooks are registered in assets/hooks.json and can:

• Log structured events
• Trigger notifications
• Update metrics
• Call external services
• Modify execution flow (rare, requires explicit enable)

---

Error Handling & Recovery

Error Categories

| Error Type | Recovery Strategy | |------------|------------------| | ValidationError | Return validation error, don't proceed | | TimeoutError | Return timeout, suggest simplification | | ToolError | Log error, try fallback, add limitation | | AgentError (LLM failure) | Retry with exponential backoff (max 3) | | ConfigError | Return configuration error, don't proceed |

Retry Policy

Retry attempt 1: immediate
Retry attempt 2: 1.5s delay
Retry attempt 3: 3s delay
After 3 failures: flag as limitation, continue or fail

---

Configuration Management

Config Resolution Order

1. Explicit keyword arguments to load_config()
2. Environment variables (ACMR_* prefix)
3. Config file (config/config.json, config/config.yaml, or $ACMR_CONFIG)
4. Built-in defaults (dataclass defaults)

Key Config Sections

• llm: Model parameters, temperature, token budgets, retry policy
• harness: Language policy, gate enforcement, degradation behavior
• knowledge_pipeline: Crawl keywords, sources, schedules, limits
• features: Feature flags for incremental rollout
• logging: Log level, output targets, rotation settings

Environment Variables

ACMR_CONFIG=/path/to/config.yaml
ACMR_LLM__MODEL=claude-sonnet-4-5
ACMR_HARNESS__DEFAULT_LANGUAGE=vi
ACMR_FEATURES__ENABLE_KNOWLEDGE_CRAWL=true

---

Knowledge Pipeline Integration

SECOND-KNOWLEDGE-BRAIN.md Structure

# Core Methods
[Domain methodology entries]

# Key Papers & References
[Tier 1/2 academic papers with DOIs]

# State of the Art
[Current research directions]

# Data Sources
[Museum collections, databases]

# Self-Update Protocol
[Crawl configuration, last updated]

# Update Log
[Append-only log of new entries]

Crawl Pipeline (tools/knowledge_updater.py)

• Sources: ArXiv (cs.GR, cs.CV, cs.AI), Semantic Scholar, Crossref, RSS feeds
• Dedup: SHA-256 of normalized DOI
• Scoring: Recency (0.4) + Keyword relevance (0.4) + Citation count (0.2)
• Schedule: Weekly academic (Mondays 8:00), Daily news (Daily 7:00)
• Safety: Backup-before-write, idempotent append, graceful degradation

---

Language Support (English/Vietnamese)

Language Detection

Pre-flight step detects language from input:

• Vietnamese: Diacritics (à, á, ả, ã, ạ, ă, â, đ, è, é, ê, ì, í, ò, ó, ô, ơ, ù, ú, ư, ý)
• English: Default
• Other: Default to English, ask user to confirm

Translation Table

All output templates support both languages:

| English | Tiếng Việt | |---------|--

…

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/ancient-costume-mural-reconstruction-agent-skill
• License: MIT

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