Manage Address Formats Setup Task Guidance logo

Manage Address Formats Setup Task Guidance

Organization
oracle
Manage Address Formats Setup Task Guidance

Exports or imports the complete Manage Address Formats setup-task data as an unfiltered Oracle Fusion FSM package, including cross-environment imports whose single approval explicitly covers a bounded source-target revision-mismatch fallback. Use only for whole-task FSM export/import; route exports through the shared task-level exporter and imports through core.data-loader validation with this task's native FSM adapter. Refuse record-level maintenance, filtered or template exports, undisclosed revision overrides, and execution that bypasses the approved route.

Overview

Publisheroracle
RepositoryFusion-Setup-Configuration-Registry
Skill nameManage Address Formats Setup Task Guidance
Stars
112
Forks
26
Bundled files
5
LicenseUPL-1.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 5 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by oracle on GitHub. Read the source before you install it.

Installation

Install the Manage Address Formats Setup Task Guidance AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/oracle/Fusion-Setup-Configuration-Registry/tree/main/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/oracle/Fusion-Setup-Configuration-Registry.git /tmp/Fusion-Setup-Configuration-Registry
mkdir -p .claude/skills
cp -r /tmp/Fusion-Setup-Configuration-Registry/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats .claude/skills/oracle-manage-address-formats-setup-task-guidance
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Manage Address Formats Setup Task Guidance in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Manage Address Formats Setup Task Guidance on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Manage Address Formats Setup Task Guidance is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Manage Address Formats Setup Task Guidance

Purpose and Boundary

Export or import the complete FSM setup package for Manage Address Formats. Fix the task identity to HZ_MANAGE_ADDRESS_FORMATS under Trading Community Foundation.

Do not use this skill for planning, standalone validation, template generation, filtered exports, or individual address-style inspection or maintenance. Validation is permitted only as mandatory preparation for an export or import.

Supported Operations

  1. export: export every setup-data object owned by the fixed task as one unfiltered, non-template FSM package through shared.setup-export.task-level-export.
  2. import: import one complete, unchanged FSM ZIP for the fixed task through mandatory core.data-loader validation and approval, then execute only through this skill's native-fsm-import as the selected native FSM adapter.

Treat a migration request as two operations. Complete and persist the source export before starting a separate target import preflight and approval.

One Consolidated Approval

Use exactly one conversational approval point for each export or import. Gather and validate every detail before asking for approval. Project creation, local task-context refresh, authentication, browser sign-in, source review, metadata checks, preview generation, Task Hook validation, and approval-envelope creation are preparation steps and require no business-operation approval.

For an import, bind the approval to an initial submission with SourceTargetDiffOkFlag=N and one conditional fallback with SourceTargetDiffOkFlag=Y only if Fusion explicitly rejects the first submission for a source-target revision mismatch. Disclose that bounded two-submission behavior in the summary. The same approval authorizes the conditional fallback because its target, task, archive bytes, hash, adapter, and execution options remain immutable.

The only approval or acknowledgement response this skill may ask the user to send is the standalone word approve, and only after the complete immutable execution summary is ready. Factual questions remain permitted only for a genuinely missing value that cannot be resolved safely from durable context. Never ask the user to reply with ready, continue, approve, or any other acknowledgement to start, finish, or resume authentication. Never ask whether the user approves opening a login page, starting the OAuth callback listener, using the configured local token cache, refreshing authentication, or running read-only preflight checks.

Authentication Is Unapproved Preparation

Run authentication as one uninterrupted preparation phase before the execution approval summary:

  1. Use the configured environment and the prescribed local, gitignored file token cache from the first authentication attempt. Reuse a valid cached token; otherwise start exactly one OAuth callback listener in a persistent execution session and open the configured identity-provider page.
  2. If the identity provider requires interactive user input, provide only an informational update such as: Complete the Oracle sign-in in the browser window; preparation will continue automatically. Do not request a reply, confirmation, approval, or sentinel word.
  3. Keep waiting on the same login process and callback. After authentication succeeds, continue environment identity, privilege, task, package, adapter, Task Hook, and preview validation automatically from the same waiting process, without a conversational handoff.
  4. Do not launch repeated login flows merely because the first listener has not produced output yet. Poll or wait on that listener until it completes or its declared timeout is reached.
  5. If authentication ultimately fails or times out, report an authentication blocker and the concrete remediation. Do not present an execution approval summary, do not ask for approve, and do not ask the user to reply ready. When authentication later becomes available, resume preparation directly.

Authentication permission prompts imposed by the host runtime are platform controls, not conversational approvals owned by this skill. Do not add a second chat approval around them.

Complete this preparation bundle first:

  1. Resolve the project, configured environment, fixed task, operation, scope, source or output path, execution options, and overwrite state from supplied inputs and durable context. Ask a factual question only when a value remains ambiguous or unsafe.
  2. If no implementation project exists, create the repository-prescribed draft project. If its local task context is missing or stale, run the project workspace initializer's artifact-only refresh. Do not create or update a live Oracle Run Book.
  3. Complete OAuth authentication and source or target identity checks under Authentication Is Unapproved Preparation. Treat interactive browser sign-in as authentication, never as a chat checkpoint or export/import approval.
  4. Finish catalog, privilege, live task, path, overwrite, package, Data Loader, Task Hook, and immutable-preview checks that apply to the selected operation.

After preparation succeeds, render the execution summary as a CommonMark bullet list with one labeled detail per item, in this order:

  • Environment: source for export or target for import.
  • Operation: export or import.
  • Task name: Manage Address Formats.
  • Task code: HZ_MANAGE_ADDRESS_FORMATS.
  • Scope: the exact whole-task scope.
  • File name: for export, the planned pattern <environment>_setup_task_csv_export_hz_manage_address_formats_<ProcessId>.zip and note that Fusion assigns ProcessId; for import, the exact ZIP basename.
  • File path: output directory plus the planned pattern for export; exact absolute ZIP path for import.
  • SHA-256: Computed after download; unavailable before approval for export; verified archive hash for import.
  • Execution adapter: the selected shared adapter.
  • Execution options: wait, polling, and applicable fallback settings.
  • Revision mismatch fallback: import only; bounded initial N and exact-mismatch-only Y behavior.
  • Side effect: the precise Fusion mutation and local artifacts produced. Do not collapse these details into prose or a table, and do not omit the file items merely because an export filename and hash are finalized after approval.

Ask the user once to reply exactly approve. That single approval authorizes only the immutable operation shown:

  • Export: one FSM export submission, polling, and result retrieval.
  • Import: one initial setupTaskCSVImports submission of the validated FSM ZIP and, only for Fusion's explicit revision-mismatch rejection, one conditional override submission of those same immutable bytes, followed by polling and result retrieval through the selected shared adapter.

Downstream approval records are internal representations of that same approval. Do not prompt again at a task-local, Task Hook, adapter, upload, initiation, or polling boundary. A change to any preview-bound value invalidates approval and requires a new prepared summary.

An approve message received before a ready immutable summary is not usable. Continue or repair preparation without asking for another acknowledgement; after preparation succeeds, present the one complete summary and make the operation's single approval request.

If a local failure provably happens before the first POST, permit one bounded recovery without another approval only when the preview fingerprint and every bound value are unchanged and audit evidence confirms no POST was attempted. If a POST was attempted or its outcome is uncertain, approval is consumed and automatic resubmission is forbidden, except for the exact revision-mismatch fallback already disclosed and bound in the import preview.

Direct Export Trigger and Resume

Treat export manage address formats task, export fsm task Manage Address Formats, and equivalent whole-task wording as a complete export request. Infer operation=export, the fixed task identity, template_only=false, export_criteria=[], wait_seconds=300, and poll_interval=10. Resolve the single configured Fusion environment and project without an approval prompt.

Run all safe preparation in the same turn. Persist the redacted ready preview:

text
project/{project_id}/artifacts/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats/current/export-approval-preview.json

The preview must have status=ready_for_approval and bind its fingerprint to:

  • project ID and source environment
  • operation, task name, and task code
  • whole-task, unfiltered, non-template scope
  • output directory and shared export adapter
  • empty export criteria
  • wait and polling options
  • exact live lookup artifact and lookup result

Compute the fingerprint from a canonical JSON representation of those bound values, excluding timestamps, status, and the fingerprint field itself. Never persist a token, secret, authorization header, or token-cache content.

Present the same values and ask exactly for approve. On the next standalone approve, load the persisted preview, recompute its fingerprint, recheck all bound values and the destination state, and immediately run the one shared export command. Do not repeat discovery or request another approval. Capture the ProcessId, poll it, retrieve FileContent or ProcessLog, verify and register the artifact, update project state, append token usage, and report the result. If no valid ready preview exists, repair preflight and present a new summary; never treat an unbound approval as permission for a POST.

Fixed Task Context

  • Pillar: CX
  • Offering: Customer Data Management
  • Offering code: ZCH_CDM
  • Functional area: Trading Community Foundation
  • Functional area code: ORA_HZ_CDM_FOUNDATION
  • Task name: Manage Address Formats
  • Task code: HZ_MANAGE_ADDRESS_FORMATS
  • CSV export/import supported: Y
  • Data Loader import mode: FSM

Never accept a task-code, offering, or functional-area override.

Whole-Task FSM Export

  1. Resolve or create the project, refresh missing local task-context artifacts, and select the configured source Fusion environment. Complete authentication, environment identity, privilege, destination-path, and overwrite checks as read-only preparation without a login approval or ready handshake.

  2. Use the shared task-name lookup without hierarchy filters:

    bash
    python3 shared/setup-export/task-level-export/scripts/find_setup_task_code.py \
      --task-name "Manage Address Formats" \
      --token-cache auto \
      --output project/{project_id}/artifacts/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats/current/setup-task-search.json \
      --json \
      --print

    Require exactly one returned row with the exact TaskName, TaskCode=HZ_MANAGE_ADDRESS_FORMATS, AvailableFlag=Y, and CSVExportImportSupportedFlag=Y. Live setup-task responses may omit offering and functional-area fields; validate hierarchy from the effective catalog and persisted task context. Fail closed on contradictory nonblank hierarchy data or zero/multiple exact rows. Never select the first search result merely because it appears first.

  3. Fix template_only=false and export_criteria=[]. Refuse every record, country, address-style, business-object, attribute, and template filter.

  4. Persist the immutable export-approval-preview.json, show the single execution summary, and ask once for approve as defined above. Do not route export preview creation through core.data-loader, its Task Hook, or any archive-verification command. An export has no ZIP or archive SHA-256 before Fusion completes it.

  5. After approval verification, invoke the shared exporter exactly once:

    bash
    python3 -u shared/setup-export/task-level-export/scripts/export_setup_task_csv.py \
      --data \
      --task-code HZ_MANAGE_ADDRESS_FORMATS \
      --offering-name "Customer Data Management" \
      --functional-area-name "Trading Community Foundation" \
      --project-id "{project_id}" \
      --output-dir project/{project_id}/exports/customer-data-management/trading-community-foundation/manage-address-formats/ \
      --token-cache auto \
      --wait-seconds 300 \
      --poll-interval 10

    Use a persistent execution session with unbuffered output. Empty initial output is not failure and must not trigger another export.

  6. Persist the returned ProcessId immediately, then poll and retrieve results only for that process. On COMPLETED or COMPLETED_WARNINGS, verify the ZIP, compute its SHA-256, register it, and update project state. On COMPLETED_ERRORS, retrieve and persist ProcessLog when available.

  7. On timeout, preserve the same ProcessId, last status, and elapsed time. Resume only that process after a later status request; never submit a replacement export automatically.

If execution ends after the POST but before the ProcessId is observable, reconcile with a read-only export-process lookup using the environment, task code, and run-start window. Resume exactly one matching process. Fail closed on zero, multiple, or unidentifiable candidates; a new POST requires a new preview and approval.

Whole-Task FSM Import Through the Data Loader-Authorized FSM Adapter

  1. Resolve or create the project and select the configured target environment. Require one explicit ZIP path; reject loose CSV files, multiple archives, directory scans, edited archives, and repacked archives.

  2. Ensure the project has a current task-context entry resolving the sequence, exact task name/code, and this skill path. Refresh only local artifacts when needed; do not create a live Oracle Run Book.

  3. Complete authentication under Authentication Is Unapproved Preparation, then target identity, privilege, route, and source-path checks before preview creation. Do not pause for a ready response after sign-in.

  4. Verify a readable ZIP containing exactly one member whose basename is ASM_SETUP_CSV_METADATA.xml. Require its single ObjectShortName to equal HZ_MANAGE_ADDRESS_FORMATS. Preserve the archive byte for byte and compute its SHA-256.

  5. Route the import through core.data-loader with the fixed task context and dataloader_import_mode=FSM. Data Loader owns source review, Task Hook validation, adapter selection, immutable preview, approval-envelope creation, and approved-payload release. It does not require its generic UCM uploader to be the execution adapter.

  6. During adapter selection, inspect the target pod's live dataLoaderDataSets/describe action metadata. Never add runID, sourceType, or another property to an ADF action unless that exact action advertises the property. For this setup-task FSM package, select cx.customer-data-management.trading-community-foundation.manage-address-formats.native-fsm-import and its setupTaskCSVImports route. Do not select core/data-loader/scripts/upload_and_initiate.py; generic Data Loader upload/initiation is not the native setup-task CSV import contract for this package.

  7. Build the Data Loader Task Hook preview over the exact immutable archive. Require it to retain executionRoute=data-loader, executionAdapter=cx.customer-data-management.trading-community-foundation.manage-address-formats.native-fsm-import, packageType=FSM, revisionMismatchPolicy=auto_override_on_exact_mismatch, sourceTargetDiffOkFlagSequence=[N,Y_on_exact_revision_mismatch], the fixed task name/code, absolute archive path, archive SHA-256, wholeTask=true, waitSeconds=180, and pollInterval=15. The adapter and revision policy are preview-bound; changing either invalidates approval.

  8. After every preflight check passes, display the one immutable-operation summary and ask once for the standalone response approve. On that message, record Task Hook approval internally, verify the archive hash again, persist the approved envelope, and release that exact payload to the selected task-local adapter without another prompt.

  9. Invoke the task-local adapter exactly once:

    bash
    python3 -u pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats/scripts/import_manage_address_formats.py \
      --zip-file "{setup_data_zip_path}" \
      --task-validation-envelope project/{project_id}/artifacts/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats/current/import-approval-envelope.json \
      --token-cache auto \
      --wait-seconds 180 \
      --poll-interval 15 \
      --output-dir project/{project_id}/artifacts/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats/imports/

    This adapter owns the bounded setupTaskCSVImports submission sequence, immediate ProcessId capture, polling, and ProcessLog retrieval. Do not call its endpoint or scripts before Data Loader releases the approved payload.

  10. The approved command sends SourceTargetDiffOkFlag=N first. If and only if Fusion returns the explicit HTTP 400 source-target revision-mismatch rejection without a ProcessId, continue automatically with SourceTargetDiffOkFlag=Y under the same approval. Permit no more than this one fallback submission. Stop on any other response, uncertain outcome, or failure, and never ask for a second approval for the preview-bound fallback.

  11. Do not upload this FSM ZIP through uploadDataLoaderFile, call initiateDataLoad, or send sourceType: FSM. Any changed target, archive bytes, hash, task context, adapter, revision policy, or execution option invalidates approval.

  12. Persist the accepted result and redacted diagnostics. If the adapter returns a ProcessId, resume or inspect only that process and never resubmit while its outcome is uncertain.

User Experience Contract

Keep the workflow calm, predictable, and easy to resume:

  1. During preparation, send short progress updates only when the stage changes: authentication, package validation, live task validation, and approval-ready. Never expose endpoint trivia, raw JSON, Base64 content, or internal rule names.
  2. Present exactly one approval summary using the labeled list defined above. Explain the revision fallback as a safety behavior, not as a second import: Fusion is tried with revision protection first; only an explicit revision mismatch permits one retry of the same package.
  3. After approval, report Import accepted as soon as Fusion returns a ProcessId. State whether the revision fallback was not needed or applied once.
  4. On completion, lead with the outcome and show only the target, task, status, ProcessId, immutable ZIP/hash, and artifact links. Put technical diagnostics after the outcome and only when they help resolve a failure.
  5. On timeout, say The import is still running; no new submission is needed, show the same ProcessId and last status, and offer the status-only continuation.
  6. On a blocked preflight, identify the single failed check and the concrete correction. Do not ask for approval until the full immutable summary is ready.
  7. On an uncertain POST response, say that no retry was made and that the next action is read-only reconciliation. Never imply that the import failed merely because a ProcessId was not observable.

Error and Retry Contract

Classify failures as input_validation, catalog_or_metadata_mismatch, auth_expired_or_invalid, authorization_or_privilege, transport_or_environment, fusion_api_error, process_timeout, process_failed, artifact_persistence, or approval_blocker.

  • Refresh a cache-managed token once after a read-only 401 or 403. A second authentication response blocks execution. During the approved adapter POST sequence, a cache-managed 401/403 may also refresh once; resume the same N or Y stage and never restart from N after reaching Y.
  • Treat HTTP 400 action error 27503 from generic initiateDataLoad as an adapter-contract defect. Stop, preserve whether UCM upload already succeeded and its content ID when available, and do not retry that generic route or upload the archive again.
  • Never retry an export POST or an import POST automatically except for the preview-approved Manage Address Formats fallback from N to Y after Fusion's exact revision-mismatch rejection. Limit that fallback to one second submission of the same immutable operation.
  • On timeout or failure, report environment, operation, task code, ProcessId or request ID, last status, elapsed time, diagnostic artifact, and one safe next action.
  • Never print or persist tokens, secrets, authorization headers, token-cache contents, full Base64 payloads, or sensitive customer data.

Result Contract

For export, report source environment, fixed task name/code, ProcessId, status, ZIP or log path, enclosure type, and ZIP SHA-256 when available. For import, return the Data Loader-authorized shared adapter result and report target, task name/code, immutable ZIP path and hash, ProcessId, status, and available process log. Lead with Import accepted, Import completed, Import completed with errors, or Import is still running. State revision fallback as not needed or applied once; never expose N/Y without the plain-language label. Record the skill run and token usage for every durable project activity.

Refusal Rules

  • Refuse individual address-style creation, editing, deletion, reordering, or inspection.
  • Refuse filtered, partial, criteria-based, and template/specification exports.
  • Refuse task-code overrides, loose CSV files, edited or repacked ZIPs, and packages that Data Loader cannot confirm as the fixed FSM task.
  • Refuse the generic uploadDataLoaderFile plus initiateDataLoad route when live action metadata does not advertise every proposed payload property.
  • Refuse direct endpoint, script, or adapter execution that bypasses the operation-specific approval workflow.

Golden Scenarios

RequestExpected route
Export all address formats from the configured environmentGather and validate all inputs, persist one export preview, ask once for approval, then run one shared FSM export and persist the ZIP, hash, and process result
Import the complete exported ZIPFinish Data Loader and Task Hook preflight, select this skill's native FSM adapter, show one immutable summary, ask once for approval, and submit the bounded initial-plus-exact-mismatch-fallback sequence
Fusion rejects import for a revision mismatchUnder the already approved bounded fallback, immediately resubmit the same immutable operation once with SourceTargetDiffOkFlag=Y, then poll the returned ProcessId without another approval
Export only one country or address styleRefuse partial scope without making a Fusion call
Create or edit an address styleRefuse record-level maintenance without making a Fusion call
Export polling times outPreserve the ProcessId and resume only that process after a later status request

Sources

  • knowledge/setup-task-catalog/effective/setupTasksAll.json
  • shared/setup-export/task-level-export/SKILL.md
  • shared/setup-import/fsm-csv-import/SKILL.md
  • core/data-loader/SKILL.md
  • https://docs.oracle.com/en/cloud/saas/applications-common/26c/farca/api-setup-task-csv-exports.html
  • https://docs.oracle.com/en/cloud/saas/applications-common/26c/farca/api-setup-task-csv-imports.html

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Manage Address Formats Setup Task Guidance AI skill do?

Exports or imports the complete Manage Address Formats setup-task data as an unfiltered Oracle Fusion FSM package, including cross-environment imports whose single approval explicitly covers a bounded source-target revision-mismatch fallback. Use only for whole-task FSM export/import; route exports through the shared task-level exporter and imports through core.data-loader validation with this task's native FSM adapter. Refuse record-level maintenance, filtered or template exports, undisclosed revision overrides, and execution that bypasses the approved route.

Why use Manage Address Formats Setup Task Guidance on TypingMind?

Because you install it once and use it with any model. Manage Address Formats Setup Task Guidance is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Manage Address Formats Setup Task Guidance in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/oracle/Fusion-Setup-Configuration-Registry/tree/main/pillars/cx/customer-data-management/trading-community-foundation/manage-address-formats. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Manage Address Formats Setup Task Guidance?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Manage Address Formats Setup Task Guidance?

As many as you like. As long as a model supports skills, you can use Manage Address Formats Setup Task Guidance with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Manage Address Formats Setup Task Guidance AI skill free?

Yes. It is published on GitHub by oracle under the UPL-1.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇