| name | content-discovery |
| description | Use when onboarding OCI users to available packs, registry entries, PromptLib spaces, ORA skills, Skills Hub / AI Skills Registry, Codex Enterprise apps/connectors, or non-pack content sources after installing OCI starter packs. |
| metadata | {"owner":"platform_org","last_updated":"2026-06-01T00:00:00.000Z"} |
Content Discovery
Help users discover usable agent content and adjacent connector surfaces after oci-dev-starter-pack and dev-starter are installed in their own profile. Start with orientation, then search. Keep recommendations separate from inventory: the starter path recommends the baseline packs plus aipack-core; other OCI and team packs are discoverable until vetted. Use oracle-ai-orientation when the user needs support routes, source links, or failure-mode guidance rather than catalog inventory.
First Response Shape
When invoked from /onboard, produce a compact map before offering discovery actions:
current setup - profile, harness, scope, active baseline packs, support packs.pack content - rules, skills, workflows, agents, prompts, MCP servers, configs, extras.tool surfaces - pack-rendered MCP servers, Codex Enterprise apps/connectors, MCP Gateway when visible.content catalogs - registered packs, non-pack content_paths, PromptLib, ORA skills, Skills Hub / AI Skills Registry.setup gaps - missing params, missing env refs, missing PromptLib key, auth/session gaps.next actions - inspect, connect, refresh, pull, install, or sync; mutation-gated when required.
Do not dump raw JSON in the user-facing response. Summarize source, status, owner/contact, and the next command to run.
Invocation Modes
source-tour - load references/source-guide.md, explain the available source surfaces, and use only /onboard evidence; do not run live catalog, PromptLib, registry fetch, browser, install, inspect, pull, or sync commands.topic-search - search the relevant live surfaces for a user-named task, tool, team, service, or content type; run only the read commands listed in this skill unless a mutation gate is approved.deep-discovery - refresh or inspect local catalogs after the user approves the exact mutation command; report what changed in the cache or profile.setup-gap - map missing keys, auth/session failures, and connector visibility gaps to the owning surface; do not recommend new content until the current setup gap is named.surface-map - classify a named content/tool area across the catalog-model lanes; report canonical owner/source, install path, local composition status, and unresolved overlap.
References
Load references/source-guide.md for /onboard source-tour mode and when the user asks what sources exist, where to start, or why a source is not handled by AIPack.
If the user asks where to get help, what to search next, how Oracle/OCI/Platform AI surfaces relate, or why setup failed, invoke oracle-ai-orientation and use its source map/failure map before expanding discovery.
Catalog Model
When a user asks how Oracle AI content surfaces relate, classify findings into three lanes instead of presenting one merged marketplace:
catalog/governance - PromptLib, Skills Hub / AI Skills Registry, Codex plugin marketplace, team registries, and owner docs answer what exists, who owns it, and what review/support posture it has.install/distribution - native PromptLib skill installs, ORA installs, Codex/Claude plugins, app connector catalogs, MCP Gateway, and AIPack installs answer how a user gets the content or tool surface into a runtime.local composition/update - AIPack profiles, pack registries, lockfile state, content_paths, sync, and update checks answer what is active locally, whether it overlaps with existing packs, and what needs refresh or approval.
Do not call AIPack the Oracle marketplace or collapse all catalogs into an AIPack registry. Say which lane each result belongs to, then recommend the lowest-friction next action for that lane.
Evidence To Collect Safely
- Use target profile, harness, scope, status summary, refs summary, installed pack list, and topic when
/onboard provides them.
- In
source-tour mode, do not collect additional live evidence; use the supplied /onboard evidence and the curated source guide.
- Run
aipack status --profile <profile> --json only when active pack/resource inventory is missing.
- Run
aipack profile refs <profile> --json only when param/env requirements are missing.
- Run
aipack registry sources --json, aipack registry list --json, or aipack search --available --kind pack --json only for a catalog/discovery request.
- Run
aipack prompt list, aipack prompt-lib list, or aipack prompt-lib search <query> only for prompt or PromptLib discovery.
- Run
ora --version-cli and ora --help before using ORA as a skills source. Run ora skills --help or ora skills list only when the local ORA version exposes a skills command; if it does not, report ORA as version-blocked and route to #devplat-ai-support.
When the user names a topic, search each relevant surface separately. Do not imply the catalogs are unified.
Never use aipack profile show --json for discovery output; it fully resolves environment-backed MCP values. If another tool output includes env values, discard the values and keep only key names and missing/present state.
Source Classes
| Source | What To Say | Primary Commands |
|---|
| Active packs | Already loaded into the target profile after sync. | aipack status --profile <profile> --json |
| Registered packs | Installable or inspectable bundles with profiles, rules, skills, workflows, MCP, configs, prompts, or extras. | aipack registry list --json, aipack search --available --kind pack --json |
| Non-pack registry entries | Skills-only or content-slice sources exposed through content_paths; they may overlap with full packs. | aipack registry list --json |
| PromptLib | Oracle prompt spaces that can be listed, searched, refreshed, or pulled into a profile after approval. | aipack prompt-lib list, aipack prompt-lib search <query> |
| ORA skills | ORA-managed skills exposed by the live ORA CLI/catalog; some may be backed by another source. | ora --version-cli, ora --help, ora skills --help, ora skills list |
| Skills Hub / AI Skills Registry | Separate Oracle skill catalogs; use direct links from the bootstrap guide or user-provided URLs. | Browser or documented registry UI |
| Codex Enterprise apps/connectors | Workspace-approved connectors that add tools or context outside AIPack, adjacent to pack-shipped MCP servers. | https://chatgpt.com/apps and the Oracle-approved app catalog |
| Team-local repos | Team-owned pack or skill sources; require owner/contact review before install. | Registry metadata, repo README, or owner docs |
| Support and troubleshooting sources | Owner-maintained Confluence/internal docs plus Slack support channels for current failure signals. | oracle-ai-orientation source map and exact-error searches |
Overlap Guard
Before recommending, installing, or adding a discovered item:
- Identify its backing source: full pack name/repo/path, non-pack repo/path, PromptLib space/id, ORA catalog/source, Skills Registry/Hub id, app connector, or unknown.
- If the same repo/path has a full pack in the registered pack registry, prefer the full pack and mark the non-pack or catalog item as a duplicate surface.
- Treat PromptLib, ORA skills, Skills Hub, and AI Skills Registry as correlation surfaces unless they expose a distinct backing repo or installable artifact with owner/contact metadata.
- Do not add or recommend a non-pack registry source from a catalog listing alone. Find the backing git source first, then check whether it is already a full pack.
- Include overlap notes in the output when the same team, domain, or repo appears in multiple surfaces.
Apps And MCP Servers
AIPack installs pack content and renders MCP server config for local harnesses. Codex Enterprise apps/connectors are connected through the ChatGPT app catalog instead; they are not installed, synced, or rendered by this pack.
When users need Slack, Outlook, SharePoint, or another approved enterprise connector, route them to https://chatgpt.com/apps and have them connect the Oracle-approved app from the catalog. If a connector is missing or unavailable, treat it as an Enterprise workspace/app-catalog issue, not an MCP install failure.
After the user connects an app, tell them to restart or reopen the Enterprise Codex/ChatGPT session before expecting the connector tools to appear. Local harness sync will not make Enterprise apps appear.
Topic Search
Use this routing when the user gives an interest area:
| User Need | Search First | Then |
|---|
| "I need a pack for my team/service" | aipack search <terms> --kind pack --json | Inspect owner/contact and install only after approval |
| "I need a skill/workflow for a task" | aipack search <terms> --json | Check active/full-pack results before non-pack and catalog surfaces |
| "I need prompts/examples" | aipack prompt list, aipack prompt-lib search <query> | Refresh PromptLib only after approval |
| "I need Slack/Outlook/SharePoint context" | Codex Enterprise app catalog | Connect the Oracle-approved app; do not install an MCP server unless no approved connector exists |
| "I need internal OCI tools" | Active MCP inventory and starter-pack server recipes | Use installing-mcp-servers for setup/auth failures |
| "I need authoritative internal docs" | oci-kb, PromptLib, registry search | Report provenance and freshness separately |
| "I need ORA skills" | ora --version-cli, ora --help, then ora skills --help if exposed | Report version-blocked ORA separately from missing AIPack content |
| "I am stuck / what should I search next?" | oracle-ai-orientation source map and common failure map | Search exact error in the owner docs/channel before recommending new content |
PromptLib Auth
If aipack prompt-lib list or aipack prompt-lib search <query> fails for auth, tell the user to create a VIEW_ONLY API key in Prompt Library (prompts.oracle.com -> Profile -> Preferences -> API Keys) and set it locally:
aipack config env set AIPACK_PROMPTLIB_API_KEY <key>
Use a WRITE key only for publishing PromptLib content. Never ask the user to paste a PromptLib key into chat.
Mutation Gates
When invoked from /onboard, inherit its mutation gate. Outside /onboard, stop and ask before running any command that mutates local caches, profiles, env config, connected apps, or pulled content:
aipack registry fetch --deepaipack pack inspect <pack> --jsonaipack prompt-lib refreshaipack prompt-lib pull <space-id> --add --profile <profile>aipack pack install <pack> --add --profile <profile>aipack pack add <pack> --profile <profile>aipack config env set <key> <value>
Output Contract
Return a concise discovery map:
installed - active in the target profile.recommended - vetted setup support such as aipack-core.discoverable - available in a registry or catalog but not starter-recommended.inspect first - plausible match that needs owner/source review.unsupported here - belongs to another delivery surface or is not pack-shaped.
For every item, include source surface, owner/contact when available, installed status, and overlap notes when the same domain appears in multiple catalogs.
