| 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". |
Cold-start Skill
Deterministic orientation for KubeDojo coding agents: issue in → API orient → minimal file reads → work.
Full ritual: scripts/prompts/cold-start.md
When to Use
- First call on a fresh agent session
- Picking up a GitHub issue (#N)
- After a long break when briefing may be stale
Steps
- Read parent task verbatim —
gh issue view N --repo kube-dojo/kube-dojo.github.io
- Run cold-start —
KUBEDOJO_ISSUE=N bash scripts/cold-start.sh
- Claim if assigned —
gh issue comment N --body "Claiming — worktree .worktrees/<name>"
- Worktree only — never commit on primary
main
- Handoff on demand — read the latest
.agent/session-state/* brief (pre-s196 history: docs/session-state/*) only when orient/briefing leave a narrative gap
Script output sections
Parse 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.
API-down
Script exits 0 with kubedojo:fallback (STATUS excerpt) + kubedojo:handoff-path. Continue with local files; do not treat API failure as a hard stop.
Do NOT cold-start when
The dispatch brief explicitly forbids shell scripts (e.g. headless content rewrites in worktrees where services-up is unnecessary).
