Plugin leaderboard

Every plugin in the catalog, ranked by how many model calls it removes.

Regenerated by CI from cee bench on a clean runner at commit ae6c7a2. Nothing on this page is written by hand; a plugin appears here by shipping a benchmark, not by being described well.
#PluginDeterminismEvents ErrorsCalls removed
1 log-retentionL1v0.1.0 Retains, preserves under legal hold, or flags a record for deletion past its retention window. 100% 4 0 11 of 11
2 payment-fraud-screenL1v0.1.0 Bands a payment by risk score into approve, step-up authentication, or decline. 100% 4 0 10 of 10
3 pr-size-gateL1v0.1.0 Gates a pull request by lines changed: auto-mergeable, request review, or split it. 100% 4 0 10 of 10
4 access-reviewL1v0.1.0 Flags an account for recertification when it has not been reviewed within the allowed window. 100% 4 0 8 of 8
5 refund-guardL1v0.1.0 Auto-approves a refund under a threshold and routes larger ones to a manager. 100% 4 0 8 of 8
6 sla-guardL1v0.1.0 Marks a support ticket as within SLA or breached based on a response-time threshold. 100% 4 0 8 of 8

What "calls removed" means

The baseline is a naive agent that calls a model once per step. Every deterministic step CEE executes is one call that baseline would have made and CEE did not, so determinism = deterministic steps / (deterministic steps + extractions) counts calls that did not happen rather than estimating tokens.

What this number is not. It is not a cost saving in currency, and it is not a comparison against a real agent implementation. There are no token counts here and no control group — both are on the roadmap and neither exists yet. A leaderboard that quietly turned call counts into dollars would be the first thing to collapse under a follow-up question.

Getting on this board

Ship a manifest and a benchmark. No Go required if your process fits the standard action library:

catalog/plugins/<name>/manifest.json    the workflow
catalog/plugins/<name>/benchmark.json   a set of {workflow_ref, context} events
catalog/index.json                      one entry pointing at both

Then cee lint to check it and cee bench to see where you land. The full walkthrough is in the contributing guide.

Raw cee bench output
rank plugin           determinism  events     errors   LLM calls eliminated vs agent
1    log-retention    100%         4          0        11 of 11
2    payment-fraud-screen 100%         4          0        10 of 10
3    pr-size-gate     100%         4          0        10 of 10
4    access-review    100%         4          0        8 of 8
5    refund-guard     100%         4          0        8 of 8
6    sla-guard        100%         4          0        8 of 8