oracle-ai-orientation

name	oracle-ai-orientation
description	Use when orienting OCI users to AI work at Oracle/OCI/Platform, troubleshooting first-run setup, or deciding where to search next across AIPack, Codex/OCA, MCP Gateway, PromptLib, Confluence, Slack, apps/connectors, and team packs.
metadata	{"owner":"platform_org","last_updated":"2026-05-20T00:00:00.000Z"}

Oracle AI Orientation

Orient OCI/Platform users and their agents to the right Oracle AI surface, failure class, and next source. This skill does not install packs, set secrets, sync harnesses, or mutate connected apps by itself.

References

Reference	Use when
oracle-ai-source-map.md	The user asks what exists, where to get help, which docs/channels matter, or where an unknown issue should be searched next
common-first-run-failures.md	AIPack, Codex/OCA, MCP, PromptLib, app/connector, registry, or bootstrap setup behaves unexpectedly

First Actions

1. Start from current evidence: profile, harness, scope, command, exact error text, visible tool/app/MCP state, and what changed recently.
2. If invoked from /onboard, keep the user-facing output compact and load only the reference sections needed for observed gaps.
3. Classify the issue into one lane before recommending action.
4. Use owner-maintained docs and source repos for durable instructions; use Slack for current support signals and owners.
5. When a fact affects a decision, cite the source surface and say whether it was verified from source/docs, inferred from diagnostics, or reported by a support channel.

Lane Classifier

Lane	Typical signal	First route
Engineering guardrails	PR, CM, release, incident, security, customer impact, compliance	AI Best Practices / policy sources; require human ownership and evidence
AIPack setup	aipack command, pack/profile/sync/registry/update behavior	aipack version, aipack status --profile <profile> --json, aipack profile refs <profile> --json, aipack doctor --profile <profile> --json
Pack content discovery	"What skills/packs/prompts exist?"	content-discovery, registry commands, PromptLib, Skills Hub, source map
Runtime/harness	Codex/OCA/ORA/OpenCode loading mismatch	oca-harness-setup and runtime config-plane reference
Tool access/MCP	Jira, Confluence, Bitbucket, DOPE, OCI, OTS, gateway tools	installing-mcp-servers, named-server probe, MCP Gateway docs
Enterprise apps/connectors	Slack, Outlook, SharePoint, ChatGPT app catalog, /apps	ChatGPT apps docs and https://chatgpt.com/apps; treat as Enterprise app catalog, not AIPack sync
Prompt sources	PromptLib auth/search/pull/publish	PromptLib docs and AIPACK_PROMPTLIB_API_KEY setup
Memory/prior context	User references a prior run, recurring failure, local convention, or known issue	Search active memory surfaces when available; verify any fix against docs/source before treating it as durable
Team context	Service runbooks, on-call flows, team repos, team pack choice	Registry owner/contact, team Confluence, team Slack, service owner
Unknown	User is stuck or asks "what is this doing?"	Load source map, search exact error, then pick one next source and one next command

Troubleshooting Contract

• Never ask the user to self-diagnose before checking the pack references and the relevant owner-maintained docs.
• Do not treat "installed" as "ready": confirm profile membership, setup values, rendered harness config, runtime loading, auth, and fresh-session state separately.
• Do not run broad live probes during first-run onboarding. Probe the named failing server, app, command, or source.
• Do not print secret values. Preserve key names, missing/present state, command shape, and redacted error text only.
• Do not recommend unvetted team packs for install. Present them as discoverable and inspect owner/contact/source before adoption.
• Do not duplicate MCP Gateway or Enterprise app wiring in AIPack. Route missing gateway/apps to their runtime/catalog owners.

Search Playbooks

For an unknown setup issue:

1. Search local pack references first: this skill, content-discovery, installing-mcp-servers, oca-harness-setup.
2. Search active memory surfaces when they are available, using the exact error or command; treat memory as a clue until it points to docs, source, or live evidence.
3. Search exact error text in Confluence and Slack.
4. Search by command and pack name, for example aipack sync old settings, oci-dev-starter-pack no packs configured, Codex apps not showing.
5. Search the owning channel from the source map or registry metadata before posting a new question.
6. If the answer comes from Slack, look for a durable doc/source link before treating it as final guidance.

For an unknown capability request:

1. Decide whether the user needs a runtime/tool, a pack/skill/workflow, a prompt, a connector app, or team-specific context.
2. Search the matching catalog only; do not imply AIPack, PromptLib, Skills Registry, apps, and MCP Gateway are unified.
3. Report overlap explicitly when the same domain appears in multiple sources.
4. Recommend install/connect/pull only after source, owner, maturity, auth, and mutation surface are clear.

Output Contract

Return:

• TL;DR - current lane, what is likely happening, and the single next check.
• Evidence - exact command/error/source surface used, value-free.
• Where to look next - 1-3 authoritative docs, source repos, or channels.
• Action - exact read-only command or mutation-gated command shape.
• Escalate when - the owner channel or doc path when the agent cannot verify locally.

Verify

• The answer names the owning lane before recommending a fix.
• First-run output stays compact unless the user asks for details.
• Unknowns route to Confluence, Slack, registry metadata, source repos, or owner docs instead of becoming guesses.
• No secret value, env-file content, or resolved profile show --json output is included.