agentprobe-testing 0.5.2__tar.gz → 0.5.3__tar.gz

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentprobe-testing
-Version: 0.5.2
+Version: 0.5.3
 Summary: Adaptive chaos-testing for LLM agents: a live model-driven Injector that reads an agent's real trajectory and decides where to break something, instead of scripting perturbations in advance.
 License: Business Source License 1.1
         

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/README.md

@@ -193,10 +193,10 @@ right on the page — no `curl` needed) or the raw API:
 ```python
 import agentprobe as ap
 
-fetched = ap.fetch_domain("joeblow-ai-agents_bd0725d86767c7481581fc41")
+fetched = ap.fetch_domain("meridian-ai-agents_bd0725d86767c7481581fc41")
 report = ap.quick_test(
     fetched.scenarios["wrong-order-among-several"],
-    lambda: ap.TargetAgent(system_prompt="You are JoeBlow's order support agent."),
+    lambda: ap.TargetAgent(system_prompt="You are Meridian's order support agent."),
     domain=fetched.domain,
 )
 print(report.render())
@@ -213,32 +213,34 @@ it just makes your test pass without proving anything real about your
 agent. Only fetch a key you generated for your own business, the same
 rule as never installing a package from someone you don't trust.
 
-## The website: accounts, personal API keys, and a dashboard
+## The website: sign in and run a real test in your browser
 
 [agentprobe-api.agentprobe.workers.dev](https://agentprobe-api.agentprobe.workers.dev)
 is the real site, not just the API — sign in with GitHub and you get:
 
-- **Your Keys** — every domain you've generated (via `/v1/generate` with
-  your API key attached), click-to-copy, so you don't have to keep your
-  own notes of which key was which business.
-- **Your Runs** — a history of test runs you've uploaded, each with its
-  full report viewable in the browser.
-- **API access** (on the Dashboard) — create a personal key
-  (`ap_live_...`, shown once at creation) to authenticate the two things
-  above from your own scripts.
-- **`/generate`** — the same domain generation as `POST /v1/generate`,
-  as a web form instead of `curl` — fill it in, get back the key and the
-  generated source to review right on the page.
-- **`/run`** — execute a real chaos test from the website itself, no
-  terminal needed: paste a domain key, pick a scenario from a real
-  dropdown (or "All scenarios" to run every one, combined into a single
-  report), point at your agent's HTTP endpoint and your own Anthropic API
-  key, and it runs in the background — the page comes back immediately so
+- **`/run`** — one flow, describing to running: describe your business in
+  plain English (same generation as `POST /v1/generate`, as a web form
+  instead of `curl`), and it writes a real test domain for it, then falls
+  straight into picking a scenario from a real dropdown (or "All
+  scenarios" to run every one, combined into a single report), naming the
+  run, and pointing at your agent's HTTP endpoint plus your own Anthropic
+  API key. It runs in the background — the page comes back immediately so
   you can go do something else on the site while it finishes, and it
   shows up under Your Runs once it does. Free to use — you're billed by
   Anthropic for your own API usage, the same as running it yourself. See
   [`runner/README.md`](runner/README.md) for how this actually executes server-side (the real `agentprobe-testing`
   package, not a reimplementation).
+- **Your Runs** — a history of test runs you've uploaded or executed,
+  each with its full report viewable in the browser, and a Rerun link
+  that jumps straight back into `/run`'s scenario/agent step for any run
+  that was started from the site.
+
+There's no separate domain-key page on the site anymore — a domain
+generated through `/run` is only ever referenced internally, not
+something you need to copy around. A personal API key
+(`ap_live_...`) for scripting the CLI/library against your account, if
+your account has one, is issued separately from the website (there's no
+self-serve "create a key" page).
 
 Attribute a generated domain to your account by passing the key as a
 bearer token:
@@ -256,7 +258,7 @@ same key to `quick_test()`:
 ```python
 report = ap.quick_test(
     scenario, my_agent, domain=fetched.domain,
-    upload_api_key="ap_live_...",       # from Your Dashboard
+    upload_api_key="ap_live_...",       # a personal key issued for your account
     domain_key=fetched_key,             # optional -- links the run back to its domain
 )
 ```

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/agentprobe/__init__.py

@@ -60,7 +60,7 @@ from agentprobe.scenario import CommitPattern, FactPattern, GoalSpec, Scenario
 from agentprobe.scenarios.registry import ALL_SCENARIOS as TICKET_SCENARIOS
 from agentprobe.scenarios.registry import BY_ID as TICKET_SCENARIOS_BY_ID
 
-__version__ = "0.5.2"
+__version__ = "0.5.3"
 
 __all__ = [
     "Agent",

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/agentprobe/quickstart.py

@@ -136,7 +136,7 @@ def quick_test(
     or a plain decide_fn(task, history) -> AgentAction, which gets wrap_agent()
     applied automatically -- either way, nothing else here changes.
 
-    Pass `upload_api_key` (a personal key from /dashboard) to also upload the
+    Pass `upload_api_key` (a personal key issued for your account) to also upload the
     finished report to your AgentProbe account, so it shows up under Your
     Runs -- see registry.upload_run for what this does and its failure
     behavior. Pass `domain_key` too if `domain` came from fetch_domain(), so

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/agentprobe/registry.py

@@ -125,11 +125,11 @@ def upload_run(
     recorded_injections: Optional[list[dict]] = None,
 ) -> None:
     """Upload a finished quick_test() report to your AgentProbe account (POST
-    /v1/runs) so it shows up under Your Runs on the dashboard. `api_key` is a
-    personal key created on /dashboard -- create one there, it's never shown
-    again after creation so save it somewhere real (an env var, not a
-    committed file). `report` must cover exactly one scenario, same as every
-    report quick_test() itself produces.
+    /v1/runs) so it shows up under Your Runs on the website. `api_key` is a
+    personal key issued for your account -- it's never shown again after
+    issuance so save it somewhere real (an env var, not a committed file).
+    `report` must cover exactly one scenario, same as every report
+    quick_test() itself produces.
 
     Pass `recorded_injections` (a list of agentprobe.injector.
     serialize_armed_injection() dicts) to make the uploaded run replayable
@@ -274,7 +274,7 @@ def replay_run(
     target_model: str = "custom",
 ) -> "Report":
     """Replay the exact chaos sequence recorded for `run_id` (an id from
-    your dashboard's Your Runs table, or a RegressionResult.previous_run_id)
+    the website's Your Runs table, or a RegressionResult.previous_run_id)
     against a new agent or model version, instead of paying for a fresh
     ModelInjector decision sequence -- only the Target's own calls repeat.
 

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/agentprobe/report.py

@@ -564,14 +564,14 @@ class Report:
     {classification_html}
   </section>
   <section>
-    <h2>By injection kind &mdash; handled rate</h2>
+    <h2>By injection kind: handled rate</h2>
     {kind_table}
   </section>
   <section>
-    <h2>Scenario detail &mdash; step by step</h2>
+    <h2>Scenario detail: step by step</h2>
     {scenario_detail_html}
   </section>
-  <footer>Generated by AgentProbe &mdash; adaptive chaos-testing for LLM agents.</footer>
+  <footer>Generated by AgentProbe. Adaptive chaos-testing for LLM agents.</footer>
 </div>
 </body></html>"""
 
@@ -621,7 +621,7 @@ class Report:
             notes = ""
             for a in traj.injections:
                 note_class = "injection-note" if a.valid else "injection-note injection-note--invalid"
-                validity = "" if a.valid else " &mdash; <strong>INVALID</strong>"
+                validity = "" if a.valid else " (<strong>INVALID</strong>)"
                 notes += (
                     f"<div class='{note_class}'>&#9889; injection fired: <strong>{esc(a.injection.kind.value)}</strong> "
                     f"@ step {a.fired_at_step} (trigger: <code>{esc(a.trigger_kind)}</code>){validity}"

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/agentprobe_testing.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: agentprobe-testing
-Version: 0.5.2
+Version: 0.5.3
 Summary: Adaptive chaos-testing for LLM agents: a live model-driven Injector that reads an agent's real trajectory and decides where to break something, instead of scripting perturbations in advance.
 License: Business Source License 1.1
         

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "agentprobe-testing"
-version = "0.5.2"
+version = "0.5.3"
 description = "Adaptive chaos-testing for LLM agents: a live model-driven Injector that reads an agent's real trajectory and decides where to break something, instead of scripting perturbations in advance."
 requires-python = ">=3.11"
 dependencies = ["anthropic>=1.0.0", "python-dotenv>=1.0.0"]

{agentprobe_testing-0.5.2 → agentprobe_testing-0.5.3}/tests/test_registry.py

@@ -175,9 +175,9 @@ def test_upload_run_includes_domain_key_when_given():
     report = quick_test(scenario, _clean1_decide)
     poster = make_poster()
 
-    upload_run(report, api_key="ap_live_test123", domain_key="joeblow_1234", poster=poster)
+    upload_run(report, api_key="ap_live_test123", domain_key="meridian_1234", poster=poster)
 
-    assert poster.calls[0]["body"]["domain_key"] == "joeblow_1234"
+    assert poster.calls[0]["body"]["domain_key"] == "meridian_1234"
 
 
 def test_upload_run_rejects_a_report_covering_more_or_fewer_than_one_scenario():