菜单
Loop through open discovery tickets until queue is empty
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
This skill should be used when the user asks to "implement X", "add X feature", "create X", "refactor X", "fix X bug", "develop this story", "work on story", or needs guidance on development work - both ad-hoc tasks and story-based workflows with style guide compliance.
Expand a ticket by linking or creating children for subsystems found in findings
Initialize project discovery — scan repo, gather intent, create initial tickets
Scan repository for structure and code ownership signals
Intent-driven discovery — recall existing knowledge or create new investigation tickets
Extract learnings from Claude Code session history.
| name | discover-loop |
| description | Loop through open discovery tickets until queue is empty |
| user-invocable | true |
| disable-model-invocation | true |
| allowed-tools | ["Bash","Agent"] |
Process all open tickets by spawning sub-agents. Each ticket gets isolated context.
Important: Run only ONE sub-agent at a time. The knowledge graph is walked sequentially — parallel agents would race on ticket selection and duplicate work.
DISCOVER="${CLAUDE_PLUGIN_ROOT}/scripts/discover.sh"
bash "$DISCOVER" ticket status
If no open tickets, report "No open tickets" and stop.
Repeat until done:
Spawn sub-agent:
Agent({
description: "Execute next discovery ticket",
subagent_type: "discover-exec-next",
prompt: "Execute the next open ticket. Return only: ok, done, or blocked."
})
Check result:
ok → Continue loopdone → Report "All tickets processed" and stopblocked → Log which ticket blocked, continue loopAfter each iteration: bash "$DISCOVER" ticket status
When loop ends, report:
Stop early if:
Report reason for early stop.