Context Quality Assessment Command

You are orchestrating parallel quality assessment of context files using the 4-dimensional framework.

Your Purpose

Batch-assess multiple contexts in parallel to:

1. Calculate quality scores (Q = 0.40·R + 0.30·C + 0.20·S + 0.10·E)
2. Determine letter grades (A-F)
3. Provide dimensional breakdown
4. Optionally update frontmatter with quality metadata

Required Knowledge

This command requires the Context Engineering skill. If not loaded, invoke centauro:context-engineering before proceeding.

Arguments

mode (optional): Assessment mode

• dry-run (default): Calculate scores, preview results, don't modify files
• apply: Calculate scores AND update frontmatter

Examples:

# Preview assessment without changes
/centauro:assess-quality

# Same as above (dry-run is default)
/centauro:assess-quality dry-run

# Apply assessment and update files
/centauro:assess-quality apply

Pre-Flight: Discovery Phase

Step 1: Find Contexts Needing Assessment

Discover all contexts in repository:

find .centauro/contexts -name "*.md" -type f

Filter logic:

For each context file, check frontmatter:

• Missing quality_score: Needs assessment
• Has quality_score but outdated (quality_last_assessed > 30 days old): Needs re-assessment
• Has recent quality_score (< 30 days): Skip (optional)

Display discovery results:

📊 Context Quality Assessment

**Repository scan:**
- Total contexts found: [N]
- Need assessment: [M] contexts
- Have recent assessment: [P] contexts

**Contexts to assess:**
1. c1-instructions/file1.md - Missing quality_score
2. c2-knowledge/file2.md - Missing quality_score
3. c4-memory/file3.md - Outdated (last assessed 2024-12-01)
...

If all contexts have recent quality scores:

✅ All contexts have recent quality assessments

All [N] contexts were assessed within the last 30 days.

Options:
- Re-assess anyway: /centauro:assess-quality apply --force
- View current quality: [show summary statistics]

If this happens, ask user if they want to proceed with re-assessment.

Step 2: Validate Mode Argument

Validate mode:

If mode provided:
  - Must be "dry-run" or "apply"
  - If invalid → Error message

If not provided:
  - Use default: "dry-run"

Display mode:

🔄 Assessment Mode: [dry-run | apply]

- dry-run: Preview quality scores without modifying files
- apply: Calculate scores AND update frontmatter

If mode is "apply", confirm with user:

⚠️ Confirmation Required

You are about to UPDATE frontmatter in [M] context files.

This will add/modify:
- quality_score
- quality_grade
- quality_dimensions
- quality_last_assessed
- quality_assessment_method

Continue? (yes/no)

Wait for user confirmation. If "no", switch to dry-run mode.

---

Phase 1: Parallel Quality Assessment

Purpose: Assess contexts using parallel agents for speed

Step 1: Determine Parallelization Strategy

Calculate agents needed based on context count:

contexts_to_assess = M

If contexts_to_assess < 5:
  → agents = 1
  → contexts_per_agent = M

Else if contexts_to_assess >= 5 and contexts_to_assess < 15:
  → agents = 3
  → contexts_per_agent = ceil(M / 3)

Else if contexts_to_assess >= 15 and contexts_to_assess < 30:
  → agents = 5
  → contexts_per_agent = ceil(M / 5)

Else if contexts_to_assess >= 30:
  → agents = 10
  → contexts_per_agent = ceil(M / 10)

Display strategy:

📐 Parallelization Strategy

- Contexts to assess: [M]
- Agents: [N]
- Contexts per agent: ~[X]
- Estimated time: ~[Y] seconds

Step 2: Partition Contexts

Distribute contexts across agents:

Agent 1: [context1, context2, ...]
Agent 2: [context4, context5, ...]
...

Balance considerations:

• Roughly equal number of contexts per agent
• Consider file sizes if known (larger files to fewer agents)
• Keep component types together when possible (easier for agent)

Step 3: Launch Parallel Assessment Agents

IMPORTANT: Use a single message with multiple Task tool calls for true parallelism.

For each agent:

Assess the quality of the following contexts using the 4-dimensional framework.

MODE: [dry-run | apply]

CONTEXTS: [
  {
    "file": "path/to/context1.md",
    "component_type": "c1-instructions"
  },
  {
    "file": "path/to/context2.md",
    "component_type": "c2-knowledge"
  },
  ...
]

For each context:
1. Read full content
2. Assess Relevance (R), Completeness (C), Consistency (S), Efficiency (E)
3. Calculate overall quality: Q = 0.40·R + 0.30·C + 0.20·S + 0.10·E
4. Determine grade (A-F)
5. Provide reasoning for each dimension
6. If MODE=apply: Update frontmatter with quality metadata

Return structured assessment for each context with:
- Dimensional scores (R, C, S, E)
- Overall score (Q) and grade
- Reasoning for each dimension
- Strengths and improvements
- Update status (if apply mode)

Launch all agents simultaneously in one message.

Example for 3 agents:

Task 1: quality-assessor (contexts 1-6)
Task 2: quality-assessor (contexts 7-12)
Task 3: quality-assessor (contexts 13-18)

Step 4: Aggregate Results

Collect all assessment results from parallel agents.

Parse each agent's output to extract:

• File paths
• Quality scores (Q, R, C, S, E)
• Grades
• Update status (if apply mode)

---

Phase 2: Results Summary

Purpose: Present comprehensive assessment results

Step 1: Display Individual Assessments

For each context (top 10 if > 10 total):

## Assessment: [filename]

**Component:** [c1-c6]
**Quality Score:** 0.XX
**Grade:** [A-F with +/-]

### Dimensional Scores
- Relevance (R): 0.XX (40% weight)
- Completeness (C): 0.XX (30% weight)
- Consistency (S): 0.XX (20% weight)
- Efficiency (E): 0.XX (10% weight)

### Strengths
- [Strength 1]
- [Strength 2]

### Areas for Improvement
- [Improvement 1]
- [Improvement 2]

---

If more than 10 contexts, show top 10 and summarize rest.

Step 2: Summary Statistics

# 📊 Quality Assessment Summary

**Assessment Complete**
- Contexts assessed: [M]
- Mode: [dry-run | apply]
- Date: YYYY-MM-DD
- Agents used: [N] (parallel)
- Time: ~[X] seconds

## Grade Distribution

- **Grade A** (0.90+): [N] contexts (XX%)
- **Grade B** (0.80-0.89): [M] contexts (XX%)
- **Grade C** (0.70-0.79): [P] contexts (XX%)
- **Grade D** (0.60-0.69): [Q] contexts (XX%)
- **Grade F** (< 0.60): [R] contexts (XX%)

Visualize distribution:

A: ████████░░ (XX%)
B: ██████████████░░ (XX%)
C: ██████░░ (XX%)
D: ██░░ (XX%)
F: ░░ (XX%)

Step 3: Average Scores

## Average Dimensional Scores

- **Overall Quality (Q):** 0.XX
- **Relevance (R):** 0.XX (40% weight)
- **Completeness (C):** 0.XX (30% weight)
- **Consistency (S):** 0.XX (20% weight)
- **Efficiency (E):** 0.XX (10% weight)

Step 4: Quality Health Assessment

## Repository Health

**Target:** 70%+ contexts should be Grade B or higher
**Current:** XX% contexts are Grade B or higher

**Status:** [✅ Healthy | ⚠️ Needs Improvement | ❌ Critical]

**Interpretation:**
- ✅ Healthy: ≥70% Grade B+
- ⚠️ Needs Improvement: 50-69% Grade B+
- ❌ Critical: <50% Grade B+

[If not healthy:]
**Recommendation:** Focus on improving [N] contexts below Grade B:
- [file1.md] - Grade [X] - [Brief issue]
- [file2.md] - Grade [X] - [Brief issue]
...

Step 5: Top and Bottom Performers

## Top Performers (Highest Quality)

1. **[filename]** - Grade A+ (0.XX)
   - Exceptional [reason]
2. **[filename]** - Grade A (0.XX)
   - Excellent [reason]
3. **[filename]** - Grade A- (0.XX)
   - [reason]

## Needs Attention (Lowest Quality)

1. **[filename]** - Grade [X] (0.XX)
   - Issues: [issue 1], [issue 2]
2. **[filename]** - Grade [X] (0.XX)
   - Issues: [issue 1], [issue 2]
3. **[filename]** - Grade [X] (0.XX)
   - Issues: [issue 1], [issue 2]

**Priority:** Address contexts with Grade C or below first.

---

Final Output

Dry-Run Mode Output

# ✅ Quality Assessment Complete (Dry-Run)

**No files were modified.** This was a preview only.

## Summary
- Contexts assessed: [M]
- Average quality: 0.XX (Grade [X])
- Grade B+ or higher: XX%
- Repository health: [Status]

## Next Steps

**To apply these assessments:**
```bash
/centauro:assess-quality apply

This will update frontmatter in all [M] contexts with:

• quality_score
• quality_grade
• quality_dimensions
• quality_last_assessed
• quality_assessment_method

To view full assessment details: [Scroll up to see individual assessments]

To improve low-quality contexts: Focus on the [N] contexts with Grade C or below listed above.

### Apply Mode Output

```markdown
# ✅ Quality Assessment Complete (Applied)

**Frontmatter updated in [M] files.**

## Summary
- Contexts assessed: [M]
- Files updated: [M]
- Average quality: 0.XX (Grade [X])
- Grade B+ or higher: XX%
- Repository health: [Status]

## Files Updated

✅ **Updated frontmatter in:**
- c1-instructions/file1.md → Quality: 0.XX (Grade B+)
- c2-knowledge/file2.md → Quality: 0.XX (Grade A-)
- c4-memory/file3.md → Quality: 0.XX (Grade B)
... ([M] total)

## Quality Metadata Added (Quality Overlay)

Each context now includes quality overlay fields added to base metadata:
```yaml
# Base metadata (remains):
estimated_quality: 0.XX  # Original estimate if present

# Quality overlay (added by assessment):
quality_score: 0.XX
quality_grade: "A-F"
quality_dimensions:
  relevance: 0.XX
  completeness: 0.XX
  consistency: 0.XX
  efficiency: 0.XX
quality_last_assessed: "YYYY-MM-DDTHH:MM:SSZ"  # ISO 8601
quality_assessment_method: "automated"

Note: The overlay pattern keeps both estimated_quality (pre-assessment) and quality_score (calculated). This tracks quality evolution over time.

Next Steps

Your contexts are now ready for curation:

/centauro:curate "your task description"

The quality gate uses estimated_quality for filtering. If not present, it falls back to quality_score.

To re-assess in the future: Re-run this command. Contexts with recent assessments (< 30 days) will be skipped.

To improve low-quality contexts: Focus on the [N] contexts with Grade C or below listed in the assessment.

---

## Error Handling

### No Contexts Found
```markdown
❌ No contexts found in .centauro/contexts/

Action:
1. Check directory exists: .centauro/contexts/
2. Initialize if needed: /centauro:setup
3. Create contexts: /centauro:build

All Contexts Recently Assessed

✅ All contexts have recent quality assessments

All [N] contexts were assessed within the last 30 days.

Current quality distribution:
- Grade A: [X] contexts
- Grade B: [Y] contexts
- Grade C: [Z] contexts

Re-assess anyway? (yes/no)

Assessment Agent Failed

❌ Assessment failed for Agent [N]

Contexts: [list of contexts this agent was assessing]
Error: [error message]

Partial results available:
- ✅ Agent 1: [M] contexts assessed
- ✅ Agent 2: [P] contexts assessed
- ❌ Agent 3: [Q] contexts FAILED

Continue with partial results? (yes/no)

Cannot Update Frontmatter

⚠️ Update failed for [N] contexts

Assessment completed successfully, but frontmatter update failed for:
- [file1.md] - [error]
- [file2.md] - [error]

Successfully updated: [M] contexts
Failed updates: [N] contexts

Action: Check file permissions or manually add quality metadata to failed files

Invalid Mode Argument

❌ Invalid mode: [value]

Mode must be "dry-run" or "apply".

Examples:
- /centauro:assess-quality dry-run  (preview only)
- /centauro:assess-quality apply    (update files)

---

Performance Expectations

Contexts	Agents	Est. Time	Speedup vs Sequential
5	1	~8s	1x (baseline)
10	3	~10s	2.4x faster
18	5	~12s	3.6x faster
30	5	~18s	4x faster
50	10	~20s	6x faster
100	10	~35s	7x faster

Note: Sequential assessment of 50 contexts would take ~120 seconds vs ~20 seconds parallel.

---

Integration

This command prepares contexts for:

• /centauro:curate - Quality gate filtering
• Quality monitoring and tracking
• Context repository health checks

Workflow:

1. Run /centauro:assess-quality apply once to establish baseline
2. Use /centauro:curate with confidence (quality metadata present)
3. Re-assess periodically (monthly) or after significant context updates
4. Track quality trends over time

---

Success Criteria

A successful quality assessment:

• ✅ Discovers all contexts needing assessment
• ✅ Launches parallel agents efficiently
• ✅ Applies 4-dimensional framework consistently
• ✅ Calculates accurate scores and grades
• ✅ Updates frontmatter correctly (if apply mode)
• ✅ Provides clear summary statistics
• ✅ Identifies contexts needing improvement
• ✅ Completes in reasonable time (< 30s for 50 contexts)

---

Key Principles

Speed Through Parallelism: Use parallel agents to assess multiple contexts simultaneously.

Safe by Default: Dry-run mode is default - preview before applying changes.

Comprehensive Reporting: Provide both individual and aggregate statistics.

Actionable Insights: Identify specific contexts needing improvement.

Consistent Framework: Apply same 4-dimensional framework across all contexts.

Health Monitoring: Track repository health against target (70% Grade B+).