agent-harness

Design benchmark runs, ablations, dataset specs, and failure-analysis artifacts.

Investigate workspace state, draft patches, run targeted validation, and package engineering artifacts.

Turn a vague request into a structured problem statement, comparison frame, recommendation, and decision-ready artifact.

Produce multi-format presentation and campaign artifacts from one task thread.

Frame analytical questions, chart directions, metrics, and reproducible data work products.

Systematically research a topic or repository using external evidence, claims validation, and synthesized outputs.

Discover relevant skill packages and supporting tools before committing to an execution path.

Plan user-facing pages, interface structure, and presentation surfaces with visible output artifacts.

Broad fallback package for open-ended tasks that need decomposition, discovery, synthesis, and a usable artifact.

Build operational playbooks, risk controls, and escalation-ready artifacts.

Ground a task in the local workspace by scanning files, extracting signals, and preparing executable next actions.