AX = Agent Experience. Measure and improve how well a tool, API, SDK, skill, or system serves an average agent — derived first-principles from what an agent fundamentally is, not from translating DX.
This project is a skill (skill/agent-ergonomics/SKILL.md)
that is both an evaluator (system → measured AX) and a designer (→
prioritized fixes + patches). Full model + rationale:
docs/ax-model.md.
- Matrix: 5 surfaces (Disclosure · Interface · Loop · Recursion · Human) × 6 orthogonal lenses (Coherence (DRY↔WET) · Economy · Determinism · Verifiability · Prior-alignment · Safety). Each cell = how that surface fares on that quality.
- Measurement: a purposeful instrument toolbox (telemetry, counters, perplexity/cold-call, Elo, IRT, trajectory, Pareto, friction-coding) — no fixed integer rubric. Combines into an AX profile + Pareto (+ optional scalar).
- Probes: a standardized battery T0–T6 + static run by a baseline
workagent(harness-agnostic) — incl. T5 compaction-survival (≥3 levels) and T6 refinement/autonomy-ceiling. - Workflow: Screen → Deep → Designer.
- Output: a
.workliterate report whose evidence is source blocks referencing real code at real lines; this project ships the.workrenderer.
Defined (via grill-me). Build order: Full Deep harness first — see the
backlog in docs/ax-model.md.
First target: the OOTA catalog. Own vendored subtree at
github.com/shinyobjectz/agent-ergonomics.