CtrlK
BlogDocsLog inGet started
Tessl Logo
PluginPublic

adobe/workfront

Adobe Workfront skills for AI agents. Includes Workfront Planning (WFP/Maestro) solution architecture: workspace design, record connections, formula fixes, tier limits, the Planning API, and a Fusion scenario optimizer.

Quality

93%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide
SecuritybySnyk

Passed

No findings from the security scan

Overview
Quality
Evals
Security
Files

automations-deep-dive.mdskills/workfront-planning-advisory/references/synthesized/

Workfront Planning automations — Deep dive and decision tree

Synthesis doc (NOT a single Adobe page). Reconciles native record-type automations + Fusion scenarios + CX Enterprise Coworker/retained AI Assistant + Planning API + request-form approvals into one decision framework. References the primary Adobe docs for each surface.

Last update: May 11, 2026 (compiled) Coworker has a phased rollout; check tenant availability. See the availability ledger in ../release-and-access.md for dates, licensing, and excluded industries.

The five automation surfaces in Workfront Planning

SurfaceTriggered byAuthored byEditable after saveAuditLatency
Native automation (button click)User click on selected recordsWorkspace Manager+Yes (except Action)Record historySynchronous, near-instant
Native automation (field value change)Field condition matchSysadmin onlyNORecord historyAsync, seconds
Fusion scenario (Watch Events trigger)Record/type/workspace create/update/deleteFusion developerYes (except webhook filters)Fusion execution history + record historyAsync, typically <1 minute
Coworker/retained AI Assistant promptUser natural-language promptAnyone with permission; Coworker writes require admin-granted accessN/A (one-shot)Record historySynchronous
Request form approvalSubmission to gated formWorkspace Manager+YesRequest status + record fieldsHuman-driven

Decision tree: when to use which

Use NATIVE BUTTON-CLICK automation when:

  • The trigger is "a user explicitly decides to do this thing now."
  • The action is one of: create project(s), portfolio, program, group, or another Planning record. Workfront object creation is unavailable in standalone Planning.
  • Permissions are stable (workspace manager + select users authorized).
  • The "Action" field will never change. (You cannot edit Action post-save.)

Use NATIVE FIELD-VALUE-CHANGE automation when:

  • The trigger is a Planning-internal state transition (e.g., Status → "Approved").
  • The same 5-condition limit and 6-action set apply. Workfront object creation is unavailable in standalone Planning.
  • You have a sysadmin available to author it.
  • You're OK with NO post-save edits. Treat as a production deploy.

Use FUSION when:

  • The trigger is anything OUTSIDE Planning (Workfront task created, external app event, scheduled time).
  • The action is anything Planning's native automations don't support (multi-step orchestration, conditional branching, cross-system updates, file generation, notifications to non-Workfront channels).
  • You need observability and replay (Fusion has execution history; native automations have minimal logging).
  • High-volume creates/updates where you need throttling control.

Use COWORKER / RETAINED AI ASSISTANT when:

  • Ad-hoc, user-driven, one-time-ish work.
  • Bulk operations that fit a natural-language description but don't justify building a permanent automation.
  • The user can verify the confirmation gate before execution.
  • NOT for repeating or scheduled work. AI Assistant and Coworker are not a cron.
  • Coworker is default read-only. Verify admin-enabled writes and tenant rollout; excluded sensitive industries retain Assistant. Planning Designer (Beta agreement required) is a separate workspace-generation surface.

Use PLANNING API directly when:

  • You're building a custom application (App Builder or external).
  • Fusion's modules don't expose the operation you need (Search records with complex $and/$or composition, batch read of 5K+ records with paginated offset in v1).
  • You need an External lookup field in a legacy Workfront custom form to display live Planning data.
  • Adobe recommends API v2; v1 remains available. See ../api-contract.md for the new version.

Use REQUEST-FORM APPROVAL when:

  • A gate is needed BETWEEN user submission and record creation.
  • Approvers may be outside Planning (license: Any).
  • The workflow is: submit → approve/reject → record (or no record).

Composability patterns

Pattern 1: Approval-gated record + auto-create child project

Request form with approval → on approval, record is created → native field-value-change automation on the new record creates downstream project. Approval applies to creation; downstream cascade is automatic. Sysadmin must own the field-value-change step.

This project bridge requires Workflow eligibility. In standalone, keep the downstream action inside Planning.

Pattern 2: External system event → Planning record

Fusion Watch Events on the external system (e.g., Salesforce, Adobe Experience Manager, GenStudio) → Fusion Create a record. Webhook filters are immutable once saved; design carefully.

Pattern 3: Planning record change → external notification

Fusion Watch Events on Planning records (state: new) → HTTP / Slack / Email module. Set "Exclude updates made by this connection" to avoid loops.

Pattern 4: Coworker/AI Assistant + native automation

User creates records via Coworker (admin-enabled writes) or retained AI Assistant prompt → record's status field triggers a field-value-change automation → projects get created. The AI + automation chain is fully audited in the record's history. Workfront project creation is unavailable in standalone Planning.

Pattern 5: API-driven bulk seed + automations OFF

For large initial loads (migration from another tool):

  1. Disable all field-value-change automations BEFORE the load.
  2. Run Planning API v2 record create or bulk create.
  3. Re-enable automations. This avoids cascading automation fires during seed (which would create thousands of unwanted child projects).

Governance rules of thumb

  • Document automations like code. Name, description, owner, last-modified date in a tracking spreadsheet (or, ironically, a Planning record type called "Automations").
  • Field-value-change automations are one-way doors. If you might want to change the Action, build it as button-click instead. Or accept that you'll delete and recreate.
  • Webhook filters are also one-way doors in Fusion. Same discipline applies.
  • AI Assistant respects per-user permissions, but actions are visible to other users immediately. If a user with broad Manage permission asks AI Assistant to "delete all records older than 90 days," the deletion is real and visible to everyone using the workspace. There's no "AI Assistant sandbox." Coworker additionally requires administrator-granted write access.
  • The 200 rpm per-user rate limit is shared across all surfaces. A user driving Fusion + AI Assistant + a custom app at the same time can collectively burn through it.
  • For destructive actions, prefer human-in-the-loop. Native automations cannot delete records (the action menu only creates). AI Assistant CAN delete but requires confirmation. Coworker is read-only by default and needs admin-granted write access. Fusion CAN delete silently — gate it with manual review (e.g., write proposed-deletion records to a "to-delete" record type and have a human approve before running the actual delete).

Common anti-patterns

Anti-pattern: Webhook fan-out without "Exclude updates made by this connection"

Scenario writes back to Planning → its own webhook fires → infinite loop. Always check the exclude toggle.

Anti-pattern: Field-value-change automation on a "Status" field with many transitions

Customer adds an automation for every status. Result: cascading project creates as records move through statuses. Either restrict to one terminal status (e.g., only "Approved") or use button-click for the rest.

Anti-pattern: Using AI Assistant or Coworker for repeating work

"Every Monday, AI Assistant, create next week's campaign records." This doesn't work — AI Assistant isn't scheduled. The same applies to Coworker prompts. Use Fusion with a Scheduler trigger.

Anti-pattern: Treating the request-form approval as a workflow engine

Customers expect: submit → approver1 reviews → approver2 reviews → revise → resubmit. Sequential multi-stage approvals and reusable templates are now supported where rolled out; stages complete in order. Do not infer iterative revise/resubmit loops from that addition. For broader workflow needs, use Fusion + status fields, or wait for broader Workfront workflow features.

Standalone boundary and business rules

Standalone cannot create/connect Workfront objects, so Pattern 1 and project-creating AI/native chains require a Workflow-enabled deployment. Business rules are manager-configured conditional edit/delete restrictions, not automation actions: no global types, and no Formula, Lookup, or Reference fields in conditions. They don't block creation and apply to existing records at their next edit or delete.

Anti-pattern: Letting Anyone-with-the-link request forms persist

The form URL leaks. External submissions land. Audit and unpublish stale forms or restrict sharing to workspace members.

Cross-references

Public documentation (find with node scripts/search.js <keywords>, then fetch the .md URL):

  • Native automations: configure automations records, create workfront objects automations
  • Fusion: fusion modules
  • AI Assistant: ai assistant
  • API: api basics
  • Approvals: approval request form

Bundled references:

  • API filter operators by field type: SKILL.md Category E
  • API v2: ../api-contract.md
  • Rollout, approvals, business rules, standalone: ../release-and-access.md

README.md

tile.json