<!-- skill-ninja-START -->
## Agent Skills (Compressed Index)
> **IMPORTANT**: Prefer skill-led reasoning over pre-training-led reasoning.
> Read the relevant SKILL.md before working on tasks covered by these skills.
### Skills Index
| Skill | Path | Description |
| ------------------------------------------------------------------------ | ------------------------ | ---------------------------------------------------------------------------------------------------- |
| [agentic-workflow-guide](.github/skills/agentic-workflow-guide/SKILL.md) | `agentic-workflow-guide` | Design, review, and improve agent workflows & agent using SSOT, SRP, Fail Fast principles. Suppor... |
<!-- skill-ninja-END -->
---
## Agents Index
### Orchestrator
| Agent | File | Description |
| ------------------------------------------------------------------ | --------------------- | -------------------------------------------------------------------------------------- |
| [saiten-orchestrator](.github/agents/saiten-orchestrator.agent.md) | `saiten-orchestrator` | Scoring orchestrator — routes user intent, delegates to sub-agents, integrates results |
### Sub-Agents (Workers)
| Agent | File | SRP Responsibility | MCP Tools |
| ------------------------------------------------------------ | ------------------ | ----------------------------------------------- | ------------------------------------------- |
| [saiten-collector](.github/agents/saiten-collector.agent.md) | `saiten-collector` | Data collection & validation from GitHub Issues | `list_submissions`, `get_submission_detail` |
| [saiten-scorer](.github/agents/saiten-scorer.agent.md) | `saiten-scorer` | AI qualitative review & score adjustment | `get_scoring_rubric`, `adjust_scores` |
| [saiten-reviewer](.github/agents/saiten-reviewer.agent.md) | `saiten-reviewer` | Score consistency review & bias detection | `get_scoring_rubric`, read scores |
| [saiten-reporter](.github/agents/saiten-reporter.agent.md) | `saiten-reporter` | Ranking report generation & presentation | `generate_ranking_report` |
| [saiten-commenter](.github/agents/saiten-commenter.agent.md) | `saiten-commenter` | GitHub Issue feedback comments (via Handoff) | `gh issue comment` |
### Workflow Patterns
**Orchestrator-Workers + Prompt Chaining + Evaluator-Optimizer + Handoff**
```
User Request
│
▼
┌──────────────────┐
│ @orchestrator │
│ (Route + Control)│
└────────┬─────────┘
│
┌────▼────┐
│@collector│ Phase 0: Collect submissions from GitHub Issues
│(Collect) │ Gate: data OK?
└────┬────┘
│
┌────▼──────────┐
│ score_all.py │ Phase A: Mechanical baseline (orchestrator runs script)
│ (Python) │ Keyword matching + repo tree analysis + checklist
└────┬──────────┘
│
┌────▼────┐
│ @scorer │ Phase B: AI qualitative review (Copilot reads submissions)
│(AI Review│ ★ THIS IS THE AI JUDGMENT STEP ★
│ 5/batch) │ Reads README/description → adjusts scores via adjust_scores()
└────┬────┘
│
┌────▼─────┐
│@reviewer │ Phase C: Consistency validation (Copilot checks fairness)
│(Validate) │ Outliers, clustering, bias detection
└────┬─────┘ Gate: PASS? → If FLAG, re-delegate to @scorer
│ ▲ │
│ └────────────────────┘ (max 2 cycles)
┌────▼─────┐
│@reporter │ Phase D: Generate ranking report
│(Report) │
└────┬─────┘
│
┌────▼─────┐
│@commenter │ [Handoff] Human-in-the-Loop (user confirms)
│(Feedback) │
└──────────┘
```
**Key: Two-Phase Scoring**
- `score_all.py` = Phase A mechanical baseline (keyword matching, repo tree, checklist)
- `@scorer` = Phase B AI Review — **Copilot agent** reads submissions qualitatively,
judges novelty/depth/quality, catches keyword gaming, adjusts via `adjust_scores()`
- `@reviewer` = Phase C — **Copilot agent** validates score consistency and fairness
### Design Principles
- **Incremental Collection**: ALWAYS fetch live Issue list from GitHub and compare
with existing data. Fetch details only for new/updated Issues. Never rely on
hardcoded Issue number lists.
- **Idempotent & Merge-Based**: Batch scripts MUST be idempotent. Existing data
is preserved via upsert; never overwrite blindly. Partial runs must not corrupt
previously collected data.
