Skip to main content

🧱 Setting Up a Template Quest Repo for Sage

Location: owl-strategy/strategy/template_quest_repo_setup.md

This document outlines the structure and content of a standardized GitHub repository template that Sage (or ClaudeCode acting as Sage) can use to begin and complete a generative bioinformatics quest. It enables reproducibility, structured memory, and long-form agentic development.

🎯 Goal​

Create a template repo that supports:

  • Agentic thinking and state tracking
  • Modular, CWL-only tooling
  • Prompt-based initialization
  • Public/private tagging for visibility

📁 Recommended Directory Structure​

quest-[name]/
├── .quest/
│   ├── meta.yaml         # Quest metadata and config
│   ├── goals.md          # Natural-language problem statement
│   ├── state.yaml        # Agent's working memory and progress log
├── CWL/                  # CWL tools/workflows (no src/ code)
│   ├── [tool].cwl
│   ├── inputs.json
│   └── Dockerfile (if needed)
├── docs/
│   ├── prompt_seed.md    # Initial prompt to activate Sage
│   ├── notes.md          # Agent or human reflections
│   └── FAQ.md            # Clarifications and reminders
├── tasks/
│   ├── tests/            # CWL conformance tests
│   ├── outputs/          # Output files or logs
│   └── report.md         # Summary of task status + next steps
└── README.md             # Overview for contributors and agents

🔖 Key Files Explained​

.quest/meta.yaml​

Defines how Sage should behave:

title: "Upload harmonized VCFs to GDC"
visibility: "public"  # or "private"
quest_type: "workflow_creation"
agent_role: "workflow engineer"
prompt_style: "stepwise, test-driven"
time_budget_minutes: 90
llm_guidance: "Start with goals.md. Update state.yaml and notes.md as you progress."

.quest/goals.md​

A natural-language mission statement outlining:

  • Desired outcome
  • Constraints
  • Edge cases or success criteria

docs/prompt_seed.md​

The first thing Sage reads. Sets up the problem clearly.

You are acting as Sage, the Promptable agent. Begin by reading the quest goal and outlining the CWL tool(s) you plan to build.

✅ Setup Instructions​

  1. Clone the template
git clone https://github.com/owlstacks/template-quest-repo.git quest-[your-title]
  1. Rename and update core files
  • Edit meta.yaml → change title, quest type, and visibility
  • Edit goals.md to describe your task
  • Set up an initial prompt_seed.md
  1. Initialize version control
cd quest-[your-title]
git init
git remote add origin [your-new-repo-url]
git push -u origin main
  1. Trigger ClaudeCode or Sage
  • Paste prompt_seed.md into ClaudeCode
  • Watch it update state.yaml, write to CWL/, and self-reflect in docs/
  1. Log Outcomes
  • Ensure tasks/report.md contains a final summary
  • Validate all CWL using tasks/tests/

🚀 Optional Enhancements​

  • Include GitHub Actions to test CWL validity on commit
  • Tag each quest with skill_set: in meta.yaml
  • Add contributor notes or review comments in docs/

✅ Summary​

The template quest repo structure gives Sage the scaffolding it needs to:

  • Reason, build, and test autonomously
  • Log internal memory and checkpoints
  • Reuse proven patterns and prompts
  • Feed data back into skill set registries for future use