菜单
KubeDojo agent session orientation. Use at the start of every fresh session or when picking up a GitHub issue. Triggers on "cold start", "orient", "session start", "issue-driven".
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。
基于 SOC 职业分类
KubeDojo orchestrator role. Drives the module queue, dispatches authors/reviewers, owns PR hygiene, owns session handoffs. Use as primary role for any standalone session on this repo. Triggers on "orchestrate", "drive the queue", "main session", "standalone session".
End-of-session handoff convention for KubeDojo. MARKDOWN handoff file (AI→AI local agent state) + live-index (.agent/STATUS.md) update — never committed to git. Triggers on "session handoff", "end of session", "wrap up session", "write handoff".
Pick the right KubeDojo agent for a task across ALL activities — write / code / review / research / mechanical, not just review. Activity × lane matrix → agent → model → dispatch command. Roster as of 2026-07-07 (cursor, codex, opus/claude-headless, agy[model-selectable, the Google lane — gemini-cli RETIRED], deepseek, grok via grok-CLI/hermes). Use before any dispatch. Triggers on "which agent", "dispatch", "route to", "who should do this".
Cross-family PR review protocol for KubeDojo. Different model family than the author per docs/review-protocol.md. For ANY agent acting as reviewer (codex, composer-2.5, agy, claude). Triggers on "review PR", "R1 review", "R2 review", "cross-family review".
Write KubeDojo curriculum modules. For ANY agent acting as the T0 author (cursor composer-2.5, codex, deepseek-v4-pro, claude headless). Use when creating modules, writing theory, exercises, quizzes. Triggers on "write module", "create module", "new module", "rewrite module".
Review KubeDojo modules against the 7-dim pedagogical rubric. For ANY agent acting as reviewer (codex, composer-2.5, agy, claude). Use when reviewing, scoring, or checking modules. Triggers on "review module", "check quality", "score module".
| name | cold-start |
| description | KubeDojo agent session orientation. Use at the start of every fresh session or when picking up a GitHub issue. Triggers on "cold start", "orient", "session start", "issue-driven". |
Deterministic orientation for KubeDojo coding agents: issue in → API orient → minimal file reads → work.
Full ritual: scripts/prompts/cold-start.md
gh issue view N --repo kube-dojo/kube-dojo.github.ioKUBEDOJO_ISSUE=N bash scripts/cold-start.shgh issue comment N --body "Claiming — worktree .worktrees/<name>"main.agent/session-state/* brief (pre-s196 history: docs/session-state/*) only when orient/briefing leave a narrative gapParse the labeled blocks from stdout:
kubedojo:orient — start here for "what to do now" (primary + up to 3 alternatives)kubedojo:briefing — full compact snapshot (actions, top_modules, blockers)kubedojo:session — latest handoff path (do not read the handoff file unless needed)kubedojo:pending-decisions — blocking Decision Cards in docs/decisions/pending/Optional: bash scripts/cold-start.sh --manifest for route discovery via /api/state/manifest.
Script exits 0 with kubedojo:fallback (STATUS excerpt) + kubedojo:handoff-path. Continue with local files; do not treat API failure as a hard stop.
The dispatch brief explicitly forbids shell scripts (e.g. headless content rewrites in worktrees where services-up is unnecessary).