Use when maintaining OCM engineering workflows across multiple repos or services and you need the canonical OCM runbooks, shepherd/release anchors, prerequisites context, or TC handoff guidance.
Installation
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Use when maintaining OCM engineering workflows across multiple repos or services and you need the canonical OCM runbooks, shepherd/release anchors, prerequisites context, or TC handoff guidance.
Provide a service/team-level map for OCM engineering work:
deploy/release via shepherd
operate and troubleshoot (oncall)
manage prerequisites stacks (customer onboarding)
keep public docs / Confluence aligned (TC handoff)
When to use me
You’re an OCM maintainer working across multiple repos (shepherd, runbooks, ops tools, service repos).
When not to use
You’re working in a single repo and the answer depends on that repo’s local conventions (commands, pinned tool versions, CI gates). Use that repo’s layered AGENTS.md and local docs first.
You need step-by-step UI clicking guidance for Console/operator tasks. Use the relevant runbook/playbook instead of relying on memory.
You’re doing generic harness setup or tool governance that isn’t OCM-specific. Prefer the aipack-core aipack-system skill or the oracle/mcp repo.
Verify
Confirm you’re operating with the repo-local rules loaded:
The repo has an AGENTS.md (or equivalent rules discovery) and it contains the repo-specific commands/versions you should follow.
If guidance conflicts, follow the most specific rule closest to the working directory.
If you changed anything in this pack (skills, agents, rules, workflows, MCP allowlists/templates):
Run aipack doctor and ensure it exits 0.
Before sharing guidance broadly:
Prefer canonical anchors already listed in this skill (runbooks, shepherd/release process, TC handoff pages) over ad-hoc links.
Failure modes
Symptom: Advice is too high-level and misses repo-specific reality (different make targets, scripts, pinned versions).
Recovery: Stop and pivot to the target repo’s AGENTS.md / README; re-derive the answer from repo-local commands and validation steps.
Symptom: You don’t have access to a referenced internal system (Confluence/Bitbucket/DevOps), or the tool surface is not enabled.
Recovery: Use the appropriate ops agent routing described under “Tool routing”, or fall back to “ask for the missing artifact/link” instead of guessing.
