radioastronomyio
OrganizationTested recipes for Claude skills and hooks — methodology documentation, failure modes, and honest assessments. Not another awesome-list.
Categories
Indexed Skills (4)
gdr-prompt
Create effective Gemini Deep Research (GDR) prompts using Negative Space Bounding (NSB) methodology. Use when user needs to write a research prompt for GDR, wants to bound an agentic research task, or mentions NSB, "deep research prompt", or research that keeps wandering off-topic. Applies the three-layer constraint architecture (Anchors → Walls → Gates) to collapse latent space before research begins.
task-cycle
Execute the development loop from checkout through merge. Use when user says "start the task", "checkout the issue", "let's work on
repo-bootstrap
Bootstrap a new repository structure after onboarding conversation. Use when user says "scaffold the repo", "create the directory structure", "let's scaffold", "hydrate the repo", "write the READMEs", or "flesh out the templates". Two phases — scaffold (create directories + placeholders) and hydrate (write content).
spec-driven-prompt
Generate bounded work unit specs for agent execution on ML01. Use when orchestrator (Claude) has completed RAV discussion and is ready to hand off a discrete coding or infrastructure task. Produces spec-driven prompts with deliverable/validation pairs that give the agent implementation freedom within clear boundaries. Specs live at /opt/agents/spec/ and produce matching worklogs at /opt/agents/work-logs/. Works with any executor (Codex, Claude Code, OpenCode, Gemini) but defaults toward Codex workflows.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.