Prepare requests for the selected AKA Execution entry point, consult Gateway or SSH guidance, and preserve complete execution output and artifacts. Use for handing prepared GPU commands, evaluation, or profiling work to AKA Execution; experiment planning and result interpretation belong to their callers.
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
Execute prepared work through the entry point supplied by this AKA run. Use its supported backend options, request format, defaults, and error behavior.
Read the run's Execution tool description, schema, selected backend, and workspace
information. Use its bound executor and configuration. For standalone
aka execution, inspect aka execution --help and supply a prepared workspace. The current legacy implementation is described in
Legacy requests.
Continue existing runs with their recorded selection and execution path. Deterministic host callers retain their existing adapters; loading this Skill does not add a model call to host execution. When the application exposes an Execution tool, hand requests to it rather than constructing another executor.
atrex-gpu-gateway Skill provided in this
optimization Agent's environment before preparing backend-specific options.
See Backend Skill references for discovery and
version handling. Transfer the relevant parameters into fields supported by
the Execution entry point. Execution owns submission, polling, cancellation,
and artifact retrieval. eval --backend atrex|sol chooses an evaluation suite;
Gateway profile / URL chooses an endpoint; neither is the Execution plugin.Backend guidance may describe a newer client than the AKA adapter supports. Check the actual Execution request surface. Report unsupported options without silently changing the endpoint, operation, evaluation path, or implementation.
Preserve the caller's prepared workspace, argv, inputs, worker environment, hardware, measurement parameters, timeouts, and required outputs. Put backend options in the entry point's configuration or request fields. Keep argv argument boundaries intact and use existing credential resolution outside saved requests.
Follow the original input allowlist, private evaluation protection, and
synchronization rules. Host environment and worker environment keep their
existing meanings. The legacy wrapper's --sync names output synchronization
paths; it is not an input manifest.
Commands in the Agate Skill describe backend operations. Arguments after
aka execution -- are worker commands for the original sandbox. Translate
according to the entry point: do not place a host-side agate ... invocation
inside worker argv. The current wrapper retains the sandbox's typed / dev mapping.
Preserve return codes, stdout, stderr, structured output, and synchronized artifacts, including existing redaction. Preserve exceptions and partial output on failure. Leave measurement interpretation and acceptance decisions to Evaluation or the original caller.
Distinguish worker execution, observation, and HTTP timeouts. Keep the original job identity when submission or observation is uncertain; use the existing execution path to confirm it instead of submitting duplicate work. Existing host and backend implementations retain their retry and recovery responsibilities. Report cancellation acceptance separately from confirmed termination. Legacy host cleanup adds no new remote termination or cross-version recovery guarantee.
No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Let AKA autonomously add, run, inspect, and revise intra-kernel timeline probes for standalone CUDA/inline PTX or CuTe DSL when ordinary benchmark, NSYS, or NCU evidence cannot answer a specific kernel-internal timing question.
The English summary is being prepared. Showing the original description.
Generate a structured implementation plan from an evidence draft. Validate paths, obtain configured independent Codex and Qoder reviews, synthesize available advice against repository evidence, preserve the draft, and produce testable acceptance criteria and validation steps.
The English summary is being prepared. Showing the original description.
Learn the target framework from enabled knowledge tools and implement a baseline GPU kernel. Use this skill to understand compute semantics, determine the target platform and framework, search reference implementations, and produce a correct V0 baseline with performance records for later profile-driven optimization.
The English summary is being prepared. Showing the original description.
Run the evidence loop of one long-horizon GPU kernel optimization episode. Use this skill to reconstruct the incumbent, profile and localize a bottleneck, research progressively, plan one coherent direction, implement and repair, validate development correctness and performance, and record every decisive experiment in the episode journal.
The English summary is being prepared. Showing the original description.
Mine a per-kernel optimization trace — a git repository capturing successive versions of one kernel being optimized — into structured, gate-validated optimization-experience records for the GPU kernel wiki. Use when asked to distill an optimization run, a kernel_opt trace directory or a version ladder into wiki records; to report what such a run actually achieved; to build, extend, re-run or validate the staging store behind those records; or to explain how a trace-derived record's number, snippet or provenance was established.
The English summary is being prepared. Showing the original description.
Choose and run ACU-only, adaptive PPU in-kernel timeline, or optional bounded joint analysis for a PPU kernel. Use for device-level bottleneck diagnosis, kernel-internal critical-path questions, or evidence that genuinely needs both; do not require all three modes.
The English summary is being prepared. Showing the original description.