Design eval harnesses — task schemas, metrics, dataset versioning, eval-as-code patterns. Use when asked to "build an eval harness", "set up eval-as-code", or "version our eval datasets".
The English summary is being prepared. Showing the original description.
See how to installReview the instructions the agent will receive before you install.
You are Evals — the LLM Evaluation Engineer on the AI Operations Team.
Establish which models, prompts, or pipeline stages the harness needs to cover, and how often it needs to run (every PR, nightly, per release).
Specify the input/output schema for each task type and the metric(s) computed for it — exact match, rubric score, model-graded, or a domain-specific metric.
Version the eval dataset alongside the code (not a spreadsheet someone edits by hand), and define the harness as a runnable, CI-invocable command with a clear pass/fail exit code.
A harness design — task schema, metric definitions, dataset versioning scheme, and how it plugs into CI.
If output exceeds the 40-line CLI budget, invoke /atlas-report with the full findings. The HTML report is the output. CLI is the receipt — box header, one-line verdict, top 3 findings, and the report path. Never dump analysis to CLI.
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.
Engineering lead — hand Apex any task and it routes internally. New features, planning, reviews, status, orientation, or system takeovers.
The English summary is being prepared. Showing the original description.
Session postmortem from local transcripts — why a run repeated work, ignored the plan, took too long, or cost more than expected. Use when asked "why did that take so long", "why was that so expensive", "what went wrong in that session", "why did the agent redo that", or when preparing a bug report about agent behavior.
The English summary is being prepared. Showing the original description.
Inspect and tune the skill-manifest gate — which of the 421 tonone skills keep their description in this project's context, and what that costs in tokens. Use when asked "why can't Claude see this skill", "show the skill gate", "how many tokens do my skills cost", "trim the skill catalogue", or "undo the skill gate".
The English summary is being prepared. Showing the original description.
Plan and scope a project — discovery, challenge assumptions, present XS-XXL depth options with token and cost estimates. Use when asked to "plan this", "scope this", "how should we build X", or when a new project/feature request comes in.
The English summary is being prepared. Showing the original description.
Scope the tonone agent roster for this project — install a curated subset of agents instead of the full 100-agent bundle. Use when "cut down the agent list", "profile for this project", "too many agents", "only need the engineering core", or after apex-stats shows a roster that's mostly unused.
The English summary is being prepared. Showing the original description.
Engineering lead reconnaissance — inventory the project before planning. Use when asked to "understand this project", "orient me on this codebase", "what's the state of the repo", "what's in progress", or before starting work on an unfamiliar codebase.
The English summary is being prepared. Showing the original description.