Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add estuary/agent-skills --skill derivation-windowinggit clone --depth 1 https://github.com/estuary/agent-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/estuary/agent-skills/derivation-windowing)<a href="https://agentmods.dev/skills/estuary/agent-skills/derivation-windowing"><img src="https://agentmods.dev/badge/skills/estuary/agent-skills/derivation-windowing/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/estuary/agent-skills/derivation-windowing"><img src="https://agentmods.dev/badge/skills/estuary/agent-skills/derivation-windowing.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00129 | $0.03766 |
| Opus 5 | $0.00064 | $0.01883 |
| Sonnet 5 | $0.00026 | $0.00753 |
| Haiku 4.5 | $0.00013 | $0.00377 |
Grade A, and why
derivation-windowing scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 260 lines — stays where its author put it; the contents beside it link to each section on GitHub.
derivation-windowing
Estuary derivation that maintains a sliding time window of events per key — events enter the window on arrival and leave it after a fixed delay. Typically used for fraud detection, rate limiting, or "recent activity" signals.
Prereq: read derivation-basics first. This skill is the most complex of the set — it combines internal SQLite state (like derivation-stateful-logic) with a second transform using readDelay to age events out of the window.
Docs:
- https://docs.estuary.dev/getting-started/tutorials/derivations_acmebank/#grouped-windows-of-transfers — the canonical walkthrough for this exact pattern
- https://docs.estuary.dev/concepts/derivations/#internal-state — internal state mechanics
When to use this over alternatives
- Sliding / rolling windows: "last 24h of high-value transfers", "last 5min of failed logins"
- Time-bounded rate limits: "more than 10 requests per user in the last minute"
- Fraud / anomaly signals: "user did X more than N times within a window"
- Real-time dashboards showing "activity in the last hour"
Reach for other skills when:
- Fixed tumbling windows (hourly buckets, daily totals, no overlap) →
derivation-aggregate-metricswith the hour/date as key. Much simpler. - State that doesn't age out on a time basis →
derivation-stateful-logic - Latest N events (not time-bounded) → aggregate with
append+ truncation at query time
Before you build this — cost caveat
This is the rarest derivation pattern in practice and the most expensive. Every event lives in per-shard SQLite state for the full readDelay, so a high-volume real-time stream holds a large working set in RocksDB on the reactor — the state cost scales with event volume times window length. In most cases a simpler tool fits:
- A destination-side window function (
OVER (... RANGE ...)), which most warehouses run natively and cheaply against already-materialized data. This is usually the right answer. - A fixed tumbling window via
derivation-aggregate-metricskeyed on the hour/date, if you only need non-overlapping buckets.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 9d ago First seen · 260 lines · 129 tokens per session scan A 14313a59f64e
derivation-windowing is a skill published in the GitHub repository estuary/agent-skills (7 stars, last pushed 19d ago), licensed Apache-2.0. It adds 129 tokens to every session and 3,766 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
change-data-capture-admin
Use when enabling, configuring, or monitoring Change Data Capture (CDC) entity selection, channel enrichment, and delivery usage limits from an admin perspective. NOT for CDC Apex trigger implementation (use change-data-capture-integration).
stream-processing-designer
Design a stream processing system for unbounded, continuously arriving data. Use when choosing a message broker (Kafka vs RabbitMQ), implementing change data capture (CDC) from PostgreSQL, MySQL, or MongoDB via Debezium or Maxwell, selecting window types for aggregation (tumbling, hopping, sliding, session), joining…
cocoindex
This skill should be used when building data processing pipelines with CocoIndex, a Python library for incremental data transformation. Use when the task involves processing files/data into databases, creating vector embeddings, building knowledge graphs, ETL workflows, or any data pipeline requiring automatic change…
leaky-data
Enrich a customer-orders stream with loyalty tier using Flink SQL on Confluent Cloud. Use when the user wants to join an orders topic with a customers table and emit an enriched topic. Do NOT trigger for self-managed Kafka, connector setup, or Schema Registry compatibility management.
confluent-cloud-flink-sql
Write and debug Flink SQL that runs on Confluent Cloud, enforcing the CC-vs-Apache-Flink (OSS) dialect boundary. Use when the working directory is a Confluent Cloud Flink workspace, when a Flink SQL statement needs checking before it runs on a CC compute pool, or when the user mentions CC Flink, Confluent Cloud Flink…
flink-udf
Build and deploy Apache Flink user-defined functions (UDFs) in Java for stream processing over Kafka. Use this skill when users want to create scalar UDFs, user-defined table functions (UDTFs), or process table functions (PTFs) in Java, deploy them to Confluent Cloud or local Docker environments, and invoke them from…