agentic-chatops

Meta-skill chaining /specify → /constitute → /plan → /tasks → spec-validation → dispatch-ready. Greenfield project from idea to parallel-dev-ready in one flow. Halts with clear remediation if any phase fails the DoD gate.

Phase C of greenfield project bootstrap. Produce constitution.md (5-15 non-negotiable articles) + steering/*.md (cross-cutting agent-wide rules). Pauses for operator review before exit.

Phase D of greenfield project bootstrap. Produce design.md + data-model.md + contracts/{openapi,asyncapi,schemas/*} + adr/NNNN-*.md. Validates contracts via swagger-cli/asyncapi/ajv before exit.

Phase B of greenfield project bootstrap. Interview the operator to produce PROJECT.json + PROJECT.md + spec/NNN-*/requirements.md (EARS-only) + acceptance/*.feature scenarios. Iterates until ears-lint.py passes.

Phase E of greenfield project bootstrap. Atomize design.md + contracts into tasks.json with files_owned + dependencies + acceptance_test + risk_score per task. Iterates until task-dag-validate passes.

Query code graph database for function callers, callees, dependencies, and dead code in CubeOS and MeshSat repos.

When multiple hosts alert simultaneously (burst), create a master YouTrack issue, run per-host triage, link children, analyze correlation, and escalate the master to Claude Code.

Cross-tier review protocol — chain of verification for REVIEW REQUEST messages from Claude Code (Tier 2). Includes 5-step checklist and REVIEW_JSON output format.

Structured error reporting format for triage failures — ERROR_CONTEXT format so next tier can resume without re-doing completed work.

Escalate a YouTrack issue to Claude Code (Tier 2) for CODE IMPLEMENTATION ONLY. Use ONLY when someone explicitly asks to implement, fix, refactor, build, or write code. Do NOT use for questions, lookups, or status checks — use youtrack-lookup for those instead.

Blocked command list for exec tool — commands that must NEVER be executed, even if asked.

Infrastructure alert triage — dedup via YT search, deep PVE/K8s investigation, auto-escalation for recurring/flapping alerts, control plane deep dive for K8s controller nodes.

Kubernetes alert triage — dedup via YT search, deep control plane investigation, auto-escalation for recurring/flapping/control-plane alerts.

Query OpenClaw's episodic memory — past triage outcomes, host patterns, alert history. Use before triage to check if you've seen this host/alert before and what happened.

Look up infrastructure devices, VMs, IPs, VLANs, and interfaces from NetBox CMDB. Use this skill FIRST for any question about what a host is, where it lives, its IP, VLAN, or role. Covers all NL and GR sites (310+ objects). Preferred over LibreNMS for device identification.

Operational knowledge base — PVE node reference, IoT cluster, GR iSCSI server quick facts for triage context.

Query past incident resolutions from the knowledge base. Use when asked about prior incidents, what fixed a host/alert before, or playbooks for similar alerts. Searches by hostname or alert rule name.

Daily proactive health scan for pre-alert conditions. Checks disk space, certificate expiry, stale YT issues, VPN status, and SeaweedFS health. User-invocable via /proactive-scan.

Look up YouTrack issues by ID or list open issues. Use this skill FIRST whenever someone mentions an issue ID (CUBEOS-72, MESHSAT-27) or asks about project status, open issues, or issue details. This is for INFORMATION — not implementation. Do NOT escalate lookup requests.