---

name: full-workflow description: Execute the complete dsRNA design workflow with human confirmation at each step

Full Workflow Skill

When to Use This Skill

Use when the user requests a complete dsRNA design workflow for a target species. Trigger phrases: "design dsRNA for {species}", "full workflow", "complete analysis"

Data Storage Structure

IMPORTANT: This workflow uses two separate directories:

1. data/ - Cached input data (reusable across runs):

   • data/{assembly}/genome.fasta - Downloaded CDS sequences
   • data/{assembly}/genome_metadata.json - Assembly info
   • data/essential_genes.json - Reference database
   • data/blast_db/ - BLAST databases

2. output/{run}/ - Analysis outputs (unique per run):

   • Created at workflow start: output/YYYYMMDD-HHMMSS-{species_slug}/
   • All analysis results, figures, and reports go here
   • Example: output/20250115-143022-drosophila_suzukii/

CRITICAL: Create Output Directory First

At the START of the workflow, create the output directory:

CopyRUN_DIR="output/$(date +%Y%m%d-%H%M%S)-{species_slug}"
mkdir -p "$RUN_DIR/figures"
echo "Created output directory: $RUN_DIR"

Replace {species_slug} with a lowercase, underscore-separated species name (e.g., drosophila_suzukii, tribolium_castaneum).

Save the $RUN_DIR path and use it throughout the workflow.

CRITICAL: Human-in-the-Loop

You MUST stop after each step and wait for user confirmation before proceeding.

After each step:

1. Present a clear summary of results
2. Show any generated figures or key data
3. Ask explicitly: "Proceed to [next step]? (yes/no/adjust)"
4. Wait for user response before continuing
5. If user says "adjust" or provides feedback, incorporate it before moving on

DO NOT run multiple steps without confirmation.

Workflow Overview

Copy┌─────────────────────┐
│ 0. Create output dir│ → output/YYYYMMDD-HHMMSS-{species}/
└────────┬────────────┘
         ▼
┌─────────────────┐
│ 1. fetch-genome │ → Download CDS to data/{assembly}/ (cached)
└────────┬────────┘     Save literature to output/{run}/
         │ ✋ CONFIRM
         ▼
┌──────────────────┐
│ 2. identify-genes│ → Save to output/{run}/essential_genes.json
└────────┬─────────┘
         │ ✋ CONFIRM
         ▼
┌─────────────────┐
│ 3. design-dsrna │ → Save to output/{run}/candidates.json
└────────┬────────┘
         │ ✋ CONFIRM
         ▼
┌─────────────────┐
│ 4. blast-screen │ → Save to output/{run}/blast_results.json
└────────┬────────┘
         │ ✋ CONFIRM
         ▼
┌──────────────────┐
│ 5. score-rank   │ → Save to output/{run}/ranked_candidates.json
└────────┬────────┘
         │ ✋ CONFIRM
         ▼
┌───────────────────┐
│ 6. generate-report│ → Save to output/{run}/report.md
└───────────────────┘

Instructions

Step 0: Create Output Directory

FIRST: Create the run output directory

CopyRUN_DIR="output/$(date +%Y%m%d-%H%M%S)-{species_slug}"
mkdir -p "$RUN_DIR/figures"
echo "Output directory: $RUN_DIR"

Save this path and use it for all analysis outputs.

---

Step 1: Fetch Genome

Read and execute: dsrna_agent/skills/fetch-genome/SKILL.md

Checkpoint Output:

Copy## ✅ Step 1 Complete: Fetch Genome

**Species:** {species_name} (TaxID: {taxid})
**Assembly:** {assembly_name}
**CDS Count:** {count} sequences
**Total Length:** {length} bp

**Literature Search:** Found {n} RNAi papers
Top mentioned genes: {gene1}, {gene2}, {gene3}

**Cached Input Data:**
- data/{assembly}/genome.fasta
- data/{assembly}/genome_metadata.json

**Analysis Output:**
- {run_dir}/literature_search.json

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
**Next:** Identify essential genes using orthology and literature evidence

Proceed to Step 2 (identify-genes)? [yes/no/adjust]

WAIT FOR USER RESPONSE

---

Step 2: Identify Essential Genes

Read and execute: dsrna_agent/skills/identify-genes/SKILL.md

Checkpoint Output:

Copy## ✅ Step 2 Complete: Identify Essential Genes

**Genes Analyzed:** {total} from genome
**Essential Genes Found:** {count} candidates

**Top 10 Candidates:**
| Rank | Gene | Function | Score | Literature Support |
|------|------|----------|-------|-------------------|
| 1 | {gene} | {function} | {score} | {papers} papers |
| ... | ... | ... | ... | ... |

**Evidence Sources:**
- Ortholog matches: {n}
- Literature support: {n} genes with published RNAi data

**Files Created:**
- {run_dir}/essential_genes.json
- {run_dir}/figures/gene_ranking.png

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
**Next:** Design dsRNA candidates for top 5 genes

Proceed to Step 3 (design-dsrna)? [yes/no/adjust]

WAIT FOR USER RESPONSE

---

Step 3: Design dsRNA Candidates

Read and execute: dsrna_agent/skills/design-dsrna/SKILL.md

Design candidates for top 5 genes (3 candidates each = 15 total)

Checkpoint Output:

Copy## ✅ Step 3 Complete: Design dsRNA Candidates

**Genes Targeted:** 5 (top essential genes)
**Candidates Designed:** 15 (3 per gene)
**Target Length:** 300 bp

**Candidate Summary:**
| Gene | Candidate | Position | GC% | Design Score |
|------|-----------|----------|-----|--------------|
| {gene} | {gene}_1 | {start}-{end} | {gc}% | {score}/5 |
| ... | ... | ... | ... | ... |

**GC Content Distribution:** {min}% - {max}% (optimal: 35-50%)

**Files Created:**
- {run_dir}/candidates.json
- {run_dir}/figures/candidate_locations.png

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
**Next:** Screen candidates for off-target matches (human, honeybee)

⚠️ This step requires BLAST databases. If not set up, run: ./setup_blast_db.sh

Proceed to Step 4 (blast-screen)? [yes/no/adjust]

WAIT FOR USER RESPONSE

---

Step 4: BLAST Screening

Read and execute: dsrna_agent/skills/blast-screen/SKILL.md

Checkpoint Output:

Copy## ✅ Step 4 Complete: Off-Target Screening

**Candidates Screened:** 15
**Databases:** Human (GRCh38), Honeybee (Amel_HAv3.1)

**Safety Results:**
| Status | Count | Threshold |
|--------|-------|-----------|
| ✅ Safe | {n} | <15 bp match |
| ⚠️ Caution | {n} | 15-18 bp match |
| ❌ Rejected | {n} | ≥19 bp match |

**Detailed Results:**
| Candidate | Human Max | Honeybee Max | Status |
|-----------|-----------|--------------|--------|
| {name} | {bp} bp | {bp} bp | {status} |
| ... | ... | ... | ... |

**Files Created:**
- {run_dir}/blast_results.json
- {run_dir}/figures/safety_heatmap.png

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
**Next:** Calculate final efficacy × safety scores

Proceed to Step 5 (score-rank)? [yes/no/adjust]

WAIT FOR USER RESPONSE

---

Step 5: Score and Rank

Read and execute: dsrna_agent/skills/score-rank/SKILL.md

Checkpoint Output:

Copy## ✅ Step 5 Complete: Score and Rank Candidates

**Scoring Formula:** Combined = Efficacy × Safety
- Efficacy: GC content (30%) + Position (20%) + No poly-N (20%) + Gene essentiality (30%)
- Safety: 1.0 (<15bp) | 0.7 (15-18bp) | 0.0 (≥19bp)

**Top 5 Candidates:**
| Rank | Candidate | Gene | Efficacy | Safety | Combined |
|------|-----------|------|----------|--------|----------|
| 1 | {name} | {gene} | {eff} | {safe} | {comb} |
| 2 | ... | ... | ... | ... | ... |
| ... | ... | ... | ... | ... | ... |

**Recommendation:** {top_candidate} targeting {gene} (score: {score})

**Files Created:**
- {run_dir}/ranked_candidates.json
- {run_dir}/figures/score_breakdown.png

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
**Next:** Generate comprehensive report

Proceed to Step 6 (generate-report)? [yes/no/adjust]

WAIT FOR USER RESPONSE

---

Step 6: Generate Report

Read and execute: dsrna_agent/skills/generate-report/SKILL.md

Final Output:

Copy## ✅ Workflow Complete: dsRNA Design for {species}

**Final Report Generated**

### Executive Summary
- **Target Species:** {species}
- **Top Recommendation:** {candidate} targeting {gene}
- **Combined Score:** {score}
- **Safety Status:** {status}

### Output Directory
`{run_dir}/`

### Files Generated
- `{run_dir}/report.md` - Full scientific report
- `{run_dir}/ranked_candidates.json` - All candidates with scores
- `{run_dir}/figures/` - Visualizations

### Quick Links
- [View Full Report]({run_dir}/report.md)
- [Download Candidates JSON]({run_dir}/ranked_candidates.json)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
**Workflow complete.** Would you like to:
- Review the full report?
- Export sequences for synthesis?
- Run additional analysis on specific candidates?

Error Handling

If any step fails:

1. Report the error clearly
2. Suggest possible fixes
3. Ask if user wants to retry or skip

Example:

Copy## ❌ Step 4 Failed: BLAST Screening

**Error:** BLAST database not found at data/blast_db/human_cds

**Solution:** Run the setup script:
\`\`\`bash
./setup_blast_db.sh
\`\`\`

Retry Step 4? [yes/skip/abort]

User Adjustments

If user provides feedback at any checkpoint:

• "adjust" → Ask what they want to change
• "skip" → Move to next step (note in report)
• "go back" → Re-run previous step
• Specific feedback → Incorporate and re-run current step