Intro

Understand your agent's decisions
Replayable infrastructure for agent investigations

Trace includes

Context inventory

Model calls

Tool outputs

Analyst agents can inspect

Checkpoint forks

Fixture requirements

Replay diffs

Trial matrix

Ranked effects

Validity labels

Platform

Capture runs, branch hypotheses, and explain what changed

isplay turns agent execution into structured evidence: context inventory, checkpoints, fixture requirements, diffs, statistics, and ranked effects.

Record prompts, context, model calls, tool proposals, tool results, artifacts, and checkpoints

Run fixture-first replays from explicit checkpoints without repeating unsafe side effects

Apply typed interventions to context, prompts, model policy, or tool outputs and compare traces

Measure first divergence, fixture dependence, low-N risk, and comparable downstream effects

Keep every conclusion bounded by replay evidence instead of unsupported causal claims

Allow AI to RCA

your agentic

systems easily

Install

npx skills add combined-ai/isplay --skill isplay-analysis

Recommended: install the skill, restart your agent app, then ask it to choose the adapter, capture a run, and analyze it.