--- name: wf-planning-fusion-optimizer description: >- Review an exported Adobe Workfront Planning + Fusion scenario blueprint (.json) for performance, speed, resource use, and API call volume, and produce a prioritized optimization self-review. Use this whenever someone shares a Fusion scenario that uses Workfront Planning (Maestro) and wants it faster, cheaper, or more reliable: slow runs, long execution time, scenario timeouts, high operations consumption, too many API calls, or rate-limit / 429 errors, especially when it works in test but breaks under load or at scale. Trigger for phrases like "make my Fusion scenario faster", "why is this Planning scenario so slow", "reduce operations or API calls", "our replication or sync scenario times out", or "review this blueprint". It is specific to the Workfront Planning connector: if a scenario does not use Planning, it says so and does not apply. Read-only; it does not modify the blueprint. metadata: category: solution-architecture license: Apache-2.0 --- # Workfront Planning + Fusion scenario optimizer ## What this does Takes one or more exported Fusion scenario blueprints (`.json`) that use the Workfront Planning (Maestro) connector and produces a prioritized, read-only review of what is hurting the scenario's performance: run time, operations consumed, and API call volume (which is what drives rate-limit 429s). It reports; it never modifies the blueprint, and it does not need live workspace access. The goal is broader than rate limits. Every module execution in Fusion is an operation, operations cost run time and quota, and loop nesting multiplies them. So the same structural fixes usually improve speed, cost, and rate-limit headroom at once. Rate-limit 429s are one symptom of the same underlying inefficiency. ## Scope: Workfront Planning only This reviewer understands the Workfront Planning connector (`workfront-maestro` modules). It is not a general Fusion optimizer. If a blueprint has no Planning modules (for example a Salesforce-to-Slack scenario), the analyzer detects that and returns a clear "not applicable" result listing the connectors it did find, rather than pretending to review it. Do not force it onto non-Planning scenarios; tell the user it is Planning-specific and, if they want, offer general Fusion advice separately without this skill. ## When to use it Use it whenever a Workfront Planning Fusion scenario is the subject and the concern is speed, run time, timeouts, operations/quota, cost, reliability, or API limits, even if the person does not say "optimize" or "429". Replication, sync, and migration scenarios benefit most because they walk hierarchies and fan out calls, but it applies to any Planning scenario. ## How to run it 1. Get the exported blueprint file(s) (Fusion scenario blueprint export, `.json`). Multiple can be reviewed together. 2. Run: ``` python scripts/analyze_blueprint.py [ ...] ``` With an optional per-request call export (CSV with Resource, Method, Status Code, and optionally Duration columns) for measured context: ``` python scripts/analyze_blueprint.py --callmix ``` Add `--json` for a machine-readable result. 3. Present the Markdown report as the deliverable, findings in the order returned (highest impact first). Do not invent numbers the script did not produce. ## How to explain the findings Anchor everything in one model: **call volume and operations are module count times how often each module runs, and iteration depth is the multiplier.** A GET is cheap at the top of a scenario and expensive several loops deep. Frame each finding in terms of what it improves: calls, operations, run time, reliability, or throughput. The report labels each finding with its "Improves:" dimension for this reason. State one thing early: **there is no bulk-read endpoint.** Search and single GET are the only ways to read; bulk exists for writes only. So raising a limit or batching writes does not fix a read-heavy scenario, only reading fewer and wider does. Lead with the read findings (F1, F2, F3) for that reason. Finding codes, their rationale, and recommended fixes are in `references/findings.md`. Read it when you need the full explanation for a code; each finding already carries a one-line recommendation in the report. ## Optional: enrich with a connected Workfront Planning MCP The static analyzer is the base and needs no connection. But if a Workfront Planning MCP is connected in the session (most customers running Planning have one), use it to make the recommendations instance-specific instead of generic. It turns opaque ids in the blueprint (`Rt69e105...`, `F6a2010...`) into real record-type and field names, and, more importantly, confirms on the live schema that each read-collapse finding is actually feasible: that the child record type has a filterable reference to its parent, and that the fields a per-record GET fetches are already carried as lookups the search returns. That is the check that otherwise needs a pasted sample. Run `python scripts/analyze_blueprint.py --ids` to list the record-type and field ids the blueprint references, then resolve them through the Planning MCP. The full workflow, and one important caveat (the MCP verifies the schema and feasibility, not the Fusion connector's output nesting), are in `references/mcp-enrichment.md`. Read it when a Planning MCP is available. This step is strictly read-only, and if no MCP is connected, skip it: the static report stands on its own. ## What to be careful about - This is a heuristic structural review of the export, not a live measurement. Say so. - Reference and record-type ids differ per workspace instance; concrete fixes must be resolved against the customer's live schema. - Recommend measuring a representative run before and after, and validating in a sandbox before promoting. - Do not promise a percentage improvement. Point to where the cost is and let a real before/after measurement produce the number. - Sleep-based throttling (P1) must be removed in sequence, after the read path is cut, or bursts can get worse. Explain the ordering rather than just saying "remove it". - With `--callmix`, shares are usually sampled: treat them as proportions.