Checks a Dynamo pull request for design fit, testing and validation, code quality, and the pull request description. Use when making a pull request to Dynamo or reviewing one.
Checks a Dynamo change for performance problems in code that runs per output token, KV block, KV event, request, connection or worker. Compares it against the cost mechanisms behind Dynamo's merged performance fixes, guards external semantics, and checks theโฆ
Ensures every piece of work in ai-dynamo/dynamo starts from a tracked issue and every pull request references one. Use when starting new work, creating or sorting its Linear issue into a feature or workstream project, opening or editing a pull request, orโฆ
Describes how a fix reaches a Dynamo release branch after code freeze - cherry-picking a merged main fix into release/X.Y.Z, landing a release-only fix, or sponsoring an external contributor's Cherry-Pick Request - with the cherry-pick Linear labels, theโฆ
Adds, updates, moves, or removes content on the Dynamo Fern docs site โ standard docs pages, catalog-driven recipe and feature-benchmark pages, examples, recipes, and translations โ keeping everything in line with the documentation style guide. Use for anyโฆ
Reports a defect in the optimization skillpack itself (the skills under .agents/skills/, the documents under agent-docs/, the role contracts under agents/, and the repository AGENTS.md files) as a GitHub issue on ai-dynamo/dynamo, using the repository'sโฆ
Drafts a candidate baseline DynamoGraphDeployment from interview requirements when no catalog recipe matches the user's model, hardware, and backend, presenting per-decision evidence for the user's confirmation. Use only from user-interviewer at interviewโฆ
Consults the repository performance rules and applicable Dynamo and engine guides to select one evidence-backed optimization proposal, then writes the generator's knowledge-consult.md reasoning record. Use after perf-analyzer completes a valid AIPerf analysisโฆ
Answers "for this model, this hardware, this GPU budget and this workload, what is the best known serving configuration, and how much do I trust it?" by walking an ordered set of recipe catalogs with provenance gates, and writes a recipe dossier recordingโฆ
Validates and normalizes raw AIPerf outputs, then evaluates valid results against target SLOs and comparable prior candidates. Use after an AIPerf Job completes to produce benchmark audit, summary, and performance analysis artifacts.
Selects and freezes a question-driven AIPerf workload, objective, load policy, and Kubernetes execution manifest for a successfully deployed Dynamo candidate. Use when a candidate needs performance characterization or a comparable measurement against aโฆ
Works with Dynamo's generated CODEOWNERS - finds out who reviews a change, fixes a failing codeowners CI check, changes review routing, or grants an external contributor area-scoped ownership. Use when the codeowners check fails on a PR, a new directory isโฆ
Runs a CI health check on an ai-dynamo/dynamo pull request โ report check status, explain why full CI has or has not triggered, root-cause failed job logs, cross-reference failures against main to separate PR-caused regressions from pre-existing flakes, andโฆ
Benchmark and profile the Dynamo frontend (dynamo.frontend HTTP + tokenizer + KV router) against mock workers (dynamo.mocker). Use when measuring frontend throughput/latency, A/B-testing a frontend change, or on-CPU/off-CPU profiling the frontend or mockโฆ
Deploys one assigned DynamoGraphDeployment and proves it with an OpenAI-compatible smoke test. Use when user-interviewer has captured the user-provided baseline DGD or hypothesis-challenger has approved a later DGD.
Adversarially reviews an evidence-backed Dynamo optimization proposal and DGD draft for comparability, duplication, attribution, correctness, feasibility, and worthwhile GPU spend. Use after hypothesis-generator writes knowledge-consult.md andโฆ
Materializes one evidence-backed proposal from a flexible knowledge-consult.md into a challenger-ready deploy-draft.yaml by applying only its selected DGD change to the current successful manifest. Use after consult-perf-knowledge writes a proposedโฆ
Launches, monitors, debugs, and collects one run-scoped AIPerf Kubernetes benchmark without changing its workload semantics. Use after perf.yaml and aiperf-config.yaml have been configured for a smoke-tested Dynamo deployment.
Synthesizes a canonical user_workload.yaml and captures the user's immutable DynamoGraphDeployment from an optimization user's initial request, attachments, and minimal follow-up interview. Use as the first skill in a new Dynamo recipe optimization run, or toโฆ
Sets up a structured debugging session for a Dynamo bug โ pull the report from a Linear ticket, GitHub issue, or pasted text, capture the environment, create a persistent worklog markdown file, and walk through rebuild, reproduction, and investigation steps.โฆ
Audit Dynamo Rust hot-path `.clone()` calls, explain which clones are removable and why, and only apply clone-removal patches when explicitly requested.
Validate that a Dynamo deployment's NIXL/UCX/NCCL interconnect is ready for disaggregated serving over RDMA/NVLink. Use after deploy-dynamo-recipe brings a deployment up (especially disagg/multi-node) to confirm the KV transport is correct; useโฆ
Start or patch Dynamo router modes and run router endpoint smoke checks. Use for round-robin, KV-aware, least-loaded, or device-aware routing setup; use deploy-dynamo-recipe for recipe deployment and troubleshoot-dynamo for failure diagnosis.
Diagnose failed or unhealthy Dynamo deployments. Use when pods, model-cache jobs, PVCs, workers, frontend/router health, endpoints, or benchmark jobs fail; use deploy-dynamo-recipe or dynamo-router-starter before this for normal bring-up.
Runs deterministic byte-parity and paired performance campaigns for Dynamo offline KV-aware replay across native-G1 vLLM and SGLang configurations, including forced scheduler-pressure and disaggregated handoff lifecycles. It is used when validating replayโฆ
Drives persistent Claude Code, Codex, or OpenCode agent sessions through a Dynamo OpenAI/Anthropic-compatible endpoint over Agent Client Protocol (ACP). Use when an agent must delegate a bounded task to another coding-agent harness running a model served byโฆ
Knowledge of Fern's built-in MDX component library (accordions, callouts, cards, steps, tabs, code blocks, API-reference snippets, and more) for authoring docs pages. Use when writing or editing a Fern `.mdx` page and deciding whether a component wouldโฆ
Create self-contained interactive HTML code-review dashboards from GitHub or GitLab pull requests, checked-out branch diffs, or supplied unified diffs, with correctness and safe-to-merge scores, smell indicators, interactive component and finding diagrams,โฆ
Knowledge of Fern's site-level navigation and structure configuration โ how a docs site is organized in `docs.yml` (and product/version `.yml` files) using sections, pages, folders, tabs, tab variants, versions, products, changelogs, and site-level settings,โฆ
Creates or updates Dynamo Enhancement Proposals as GitHub issues, including lightweight DEPs, implementation plans, and retroactive DEPs for ai-dynamo/dynamo. Use when proposing a feature or architecture change, drafting an implementation plan, or filing aโฆ
Checks Dynamo Enhancement Proposal issue status, lists DEPs by lifecycle state or area, and finds related DEP issues in ai-dynamo/dynamo. Use when asking where a DEP stands or which DEPs exist for an area.
Updates Dynamo Enhancement Proposal lifecycle state in GitHub, including triage, PIC assignment, review, approval, and status label changes. Use when triaging a new DEP, assigning a PIC, or moving a proposal through review and approval.
Reviews code changes in the style of Graham King's ai-dynamo/dynamo reviews โ exacting Rust and systems-level standards covering error handling, tracing discipline, unnecessary clones, async and concurrency correctness, log levels, and minimal diff surface.โฆ