artifact-coherence-check

name	artifact-coherence-check
description	Cross-artifact consistency validation for prompt engineering systems: rule consistency, reference integrity, handoff chain validation, redundancy detection, and token budget auditing. Use when reviewing PE artifact coherence, detecting contradictions between artifacts, or auditing system-wide consistency.

Artifact Coherence Check Skill

Purpose

Validate that prompt engineering artifacts cooperate correctly without contradictions, broken references, or redundant content. Provides systematic checks that can be invoked by meta-validator (Ecosystem Audit mode), validation prompts, or directly by users.

When to Use

Activate this skill when:

• System audit: "Check if PE artifacts are consistent"
• After changes: "Verify no contradictions after updating context files"
• Dependency check: "Are all cross-references valid?"
• Redundancy scan: "Find duplicated content across PE artifacts"
• Token audit: "Check which files exceed their budget"
• Handoff validation: "Do all handoff targets exist?"

Do NOT use this skill for:

• Single-file prompt/agent validation (use prompt-engineering-validation skill)
• Article content review (use article-review skill)
• Creating or modifying PE artifacts (use builder/updater agents)

Quick Reference

Coherence Check Sequence

1. Reference Integrity  → all links resolve
2. Rule Consistency     → no contradictions between files
3. Handoff Chain        → all targets exist
4. Redundancy Scan      → no duplicated canonical content
5. Token Budget         → all files within limits
6. Dependency Map       → map matches actual state

Severity Classification

Severity	Examples	Action
CRITICAL	Contradictory rules, broken handoff chain, plan+write tools	BLOCK — must fix immediately
HIGH	Broken reference links, missing required sections	Fix before next use
MEDIUM	Token budget exceeded, minor redundancy	Fix when convenient
LOW	Stale version history, cosmetic inconsistencies	Note for future

Detailed Workflows

Workflow 1: Reference Integrity Check

Verify all cross-references between PE artifacts resolve to existing files.

Process:

1. Scan target file(s) for markdown links and 📖 references
2. Extract all referenced file paths
3. Verify each path resolves using file_search or list_dir
4. Report broken references with file, line number, and target path

Check template: reference-integrity.template.md

Common reference patterns to check:

• 📖 Complete guidance: links to context folders
• 📖 See: links to specific context files
• Template references: .github/templates/*.template.md
• Skill references: prompt-engineering-validation skill name
• Handoff targets: agent: values in YAML handoffs: arrays

Workflow 2: Rule Consistency Check

Verify rules don't contradict across artifact layers.

Process:

1. Identify canonical rules from context files (Layer 1)
2. Check instruction files (Layer 2) — do they agree or contradict?
3. Check agents (Layer 3) — do inline rules match context + instructions?
4. Check prompts (Layer 4) — do workflow rules align?

Key rules to cross-check:

Rule	Canonical Source	Check Against
Tool alignment (plan=read-only)	01.04-tool-composition-guide.md	All agents, PE-validation skill
Template-first (>10 lines)	01.01-context-engineering-principles.md (P8)	prompts.instructions.md, agents.instructions.md
Three-tier boundaries (min items)	01.06-system-parameters.md	All agents
Token budgets	01.06-system-parameters.md	All instruction files
Validation caching (7-day)	04.01-validation-caching-pattern.md	Validation prompts
Handoff send: strategy	02.01-handoffs-pattern.md	All agents with handoffs

Contradiction detection:

• Same concept with different thresholds (e.g., "5 lines" vs "10 lines")
• Different tool lists for same agent mode
• Conflicting boundary rules (one file says "Always Do", another says "Ask First")

Workflow 3: Handoff Chain Validation

Verify all agent handoff chains are complete and valid.

Process:

1. Extract handoffs: from all agent YAML frontmatter
2. For each handoff target, verify the agent file exists
3. Verify the target agent has appropriate mode and tools
4. Check for orphaned agents (exist but never referenced as handoff target)
5. Check for circular handoffs (A→B→A without termination condition)

Expected chains:

prompt-researcher → prompt-builder → prompt-validator ↔ prompt-builder
agent-researcher → agent-builder → agent-validator ↔ agent-builder
context-builder → prompt-validator
instruction-builder → prompt-validator
meta-validator → meta-optimizer → prompt-validator

Workflow 4: Redundancy Scan

Detect duplicated content that violates single-source-of-truth principle.

Process:

1. For each canonical rule (from context files), search for inline copies in agents/instructions
2. Flag any instruction/agent embedding >5 lines of content available in context files
3. Check for identical tables, checklists, or code examples across files
4. Calculate estimated token waste from redundancy

Search patterns:

Content	Canonical Source	Search Agents/Instructions For
Tool alignment rules	01.04-tool-composition-guide.md	"plan.*read-only", "write tools", "tool clash"
Boundary pattern	01.01-context-engineering-principles.md	"Always Do.*Ask First.*Never Do"
Template-first rules	01.01-context-engineering-principles.md	"10 lines", "externalize", "template-first"
Reliability checksum	02.01-handoffs-pattern.md	"Goal Preservation", "Scope Boundaries"

Workflow 5: Token Budget Audit

Verify all PE artifacts are within their token budgets.

Process:

1. Count lines for each artifact (lines × 6 ≈ tokens estimate)
2. Compare against budgets from 01.06-system-parameters.md
3. Flag files exceeding WARNING threshold
4. Recommend split or compression for files exceeding CRITICAL threshold

📖 Budget thresholds (per-type limits, warning/critical levels): 01.06-system-parameters.md → Token Budgets

Workflow 6: Dependency Map Verification

Verify the dependency map matches the actual artifact state.

Process:

1. Load the dependency-tracking file from .copilot/context/00.00-prompt-engineering/ (see 00.00-context-structure-index.md → Functional Categories)
2. Compare listed files against actual files on disk
3. Verify reference counts by spot-checking 3–5 high-impact entries
4. Flag any files present on disk but missing from the map

Templates

• Reference Integrity — Structured format for reporting broken references
• Coherence Report — Summary template for full coherence audit results

Common Issues

Issue: Contradictory Token Budgets

Symptom: Different budget numbers for same artifact type across files Solution: Update all references to match canonical source (01.01-context-engineering-principles.md)

Issue: Orphaned Agents

Symptom: Agent file exists but no prompt or other agent references it as a handoff target Solution: Either add the agent to an orchestration workflow or deprecate it

Issue: Stale Dependency Map

Symptom: New artifacts exist but aren't listed in the dependency-tracking file (see 00.00-context-structure-index.md → Functional Categories) Solution: Update the dependency map with new entries and reference counts

Resources

• 📖 Dependency map: see dependency-tracking in .copilot/context/00.00-prompt-engineering/ (00.00-context-structure-index.md → Functional Categories)
• 📖 Lifecycle management: .copilot/context/00.00-prompt-engineering/05.02-artifact-lifecycle-management.md
• 📖 Entry points: .copilot/context/00.00-prompt-engineering/05.03-pe-workflow-entry-points.md
• 📖 Context engineering principles: .copilot/context/00.00-prompt-engineering/01.01-context-engineering-principles.md