The agent experience toolkit
It passed.
But was it
usable?
AJX puts a fresh agent in your product. See where it hesitates, what breaks, and exactly what your team can improve.
Codex · Claude Code · Kiro CLI
$ demo-export export --out result.csv
error: unknown flag --out!
$ demo-export export --help--output <path> Write CSV to file
$ demo-export export --output result.csv
✓ CSV valid · 100 rowsThe task succeeded.
The first attempt didn’t.
Abbreviated display of the included fictional trace.
Top-level help gives no selection rule.
Command help reveals a working alternative.
The agent cannot distinguish work from a stall.
A clear row count confirms the result.
Make the quick-start
command executable.
Replace --out with --output and include a working CSV
example.
The documented command produces a 100-row CSV in a fresh workspace.
A pass is an outcome.
Usability is the path to it.
01 / Instrument the experience
From a real task
to a concrete next step.
Use the harness and model your customers use. Keep the task and success check explicit. Then inspect what actually happened.
Set the conditions.
Task, harness, model, access, and environment. Start with one bounded attempt.
Capture the journey.
A fresh worker tries the task. Inspect progress and artifacts in the local console.
Check the result.
Keep what the agent claimed separate from what your success check confirms.
Act on the evidence.
Prioritized asks point back to events, so your team can inspect the reasoning.
02 / Keep the comparison honest
Same task.
Different experiences.
Compare asks and outcomes across product versions and harnesses. Open the complete matrix to inspect provenance and measurement limits.
Explore the complete example ↗| What to inspect | BaselineProduct v1.2.0Codex · synthetic fixture | RevisedProduct v1.3.0Codex · synthetic fixture | HeadlessProduct v1.3.0Kiro CLI · synthetic fixture |
|---|---|---|---|
| Product asks |
3 asks
Fix the quick-start flag. |
1 ask
Show export progress. |
1 ask
Check headless authorization before export. |
| Verified outcome | ✓ Succeeded | ✓ Succeeded | ! Failed: approval required |
| Elapsed time | 1m 48s | 34s | 25s |
| Recorded tool calls Partial coverage | 5 / 1 failed | 3 / 0 failed | 2 / 1 failed |
| Inspect evidence | Open report ↗ | Open report ↗ | Open report ↗ |
All events, product versions, and measurements above are invented to demonstrate AJX reports. These runs do not establish performance differences between products, models, or harnesses.
03 / Run your first experiment
Give your agent
the starting point.
Open your project and paste the prompt. Your agent installs the native AJX CLI and skill, checks setup, and helps you define one useful test.
A project to try
A CLI, SDK, API, or developer workflow.
A signed-in worker CLI
Codex, Claude Code, or Kiro CLI on
macOS or Linux.
A quick review before the run
Agree on success, access, and
model costs.
Your results: a report, visual journey, and underlying artifacts. Once the task is
stable, add matrix runs and ajx ci acceptance checks.
Install the latest stable AJX CLI and skill by following https://raw.githubusercontent.com/royosherove/ajx/main/docs/AGENT-QUICKSTART.md Then load the installed AJX skill. I want to use AJX to test jq. Starting point: https://jqlang.org/manual/ Task: Filter a sample JSON inventory to in-stock items and export their names and prices as CSV. Prepare any tools, generated sample data, and local test server needed in a temporary workspace. Use the AJX skill to run one trial with a fresh agent, verify the result, and show me the report and journey.
Prefer a terminal? Install the CLI + skill yourself.
bash -o pipefail -c 'curl -fsSL https://github.com/royosherove/ajx/releases/latest/download/install.sh | bash' && export PATH="$HOME/.local/bin:$PATH"
macOS or Linux. The installer includes the skill. Your agent can read it directly; start a fresh session if the host needs one to discover /ajx. No GitHub login required.
One installation. CLI + Agent Skill. No Go toolchain or GitHub login.