capture-issues
Installation
SKILL.md
Capture issues
Turn whatever the user handed you into GitHub issues in the current repo (gh repo view). Ask only if the target repo or project is ambiguous.
Rules
- Fewer issues, not more. One issue per coherent unit of work. Never split what a single implementer would finish in one pass. When the input spans several themes, group the issues under milestones.
- Write for a weaker agent. Issues are implemented by coding agents less capable than you, with no access to this conversation. An issue is finished when one of them can implement it without guessing: intended architecture, files and modules to touch, data shapes and signatures, worked examples, edge cases, and acceptance criteria. Drop that depth only when the user explicitly asks for terse issues.
- Draft with the strongest model available. Prefer Fable (
model: "fable") or GPT Astra through the Codex CLI for writing the issue bodies. Never mention the model in an issue. - Create directly. Create the issues with
gh issue create(andgh apifor milestones) without asking for confirmation.
Finish with the created issue numbers, titles, links, and milestones.