@engineeros/connector 1.58.2 → 1.59.0

package/README.md

@@ -346,6 +346,14 @@ Codex connectors advertise `gpt-5.6-sol` and `gpt-5.6-terra` by default. Overrid
 ENGINEEROS_AGENT_MODELS="model-a,model-b" npx @engineeros/connector start --workspace .
 ```
 
+### One model per session, stepping up after failures, and the image probe
+
+Writing tests and repairing them, and hunting and fixing case files, run on **one model per session**: a repair or a correction resumes the session that wrote, on the same model, so the prompt cache is not thrown away. When repairs do not work (two failed repairs of Automate tests, or case files that still cannot run after the correction turn), one more attempt runs as the `escalated` class: one effort step above the choice it follows (or the choice EngineerOS sent for it), in a **fresh session** that is given the whole task and what went wrong. Its tokens are reported as their own part (`task: "escalated"`).
+
+Reading production signals (`monitoring`) runs on the engine default unless EngineerOS chose, never on the reviewer's model.
+
+The connector reports the input kinds a model accepts (`input_modalities`) when the engine says so, and tells EngineerOS when a review's screenshots were refused (`capability_failure: "images"`). When a model appears that nothing describes, EngineerOS may ask for a **vision probe** (`model.probe`, capability `model_probe`), only while this connector is idle. The probe writes two of three fixed tiny PNG pictures, each a six-digit code, to a temporary folder **outside the checkout**, and runs one read-only turn per picture asking for the code (about 3 to 5 thousand tokens in all). A wrong or refusing answer means "no images", two right answers mean "images", and an error means "unknown", never "no". The folder is removed afterwards. A person can turn probes and evals off for a connector in its routing settings.
+
 ## Codex CLI compatibility
 
 The connector prints the exact Codex CLI version it will use before connecting. If the
@@ -556,6 +564,7 @@ On Linux, missing Chromium system libraries require administrator access. At a t
 - **Locale, time zone and network**: `set locale` and `set time zone` shape the browser sessions a case starts after them. `set network` makes the current session `offline`, `online` or `slow` (Chromium throttling).
 - **Mobile web, budgets and jobs**: `set device` gives the sessions a case starts a phone's or tablet's screen, touch and browser (any device name Playwright knows), and `tap` touches an element. `responds within`, `loads within` and `largest contentful paint within` check timings. `poll` repeats a request every two seconds until its check holds, for background jobs.
 - **Local instance**: when the hunter has a launch command, the connector starts it from this checkout before the session. It waits up to three minutes for the environment's address to answer, and stops the whole process tree when the session ends.
+- **Inputs**: a hunt reads what the team gave its environment, on this machine: uploaded documents and designs (cached under `<inputs>/uploaded/`), docs, help-centre and changelog pages (guarded GETs that refuse private addresses, summarised once per change by one agent turn), an OpenAPI or GraphQL schema (checks are generated from it), a test inbox (Mailosaur, MailSlurp, Mailpit or IMAP) and a 2FA secret (`{totp:NAME}`), a reset URL, a per-host request rate (default 10 a second) and, once a person confirmed payment testing, a Stripe test-card guard that stops live keys. Secrets stay here: `npx --yes @engineeros/connector@latest secret set NAME --project PROJECT_UUID` reads the value from a hidden prompt or standard input, never an argument; `secret list` shows names only.
 
 ## Bug Hunter app surfaces and production watch
 
@@ -623,3 +632,5 @@ triaged and filed through Findings. The connector does not run ZAP; CI owns the
 ## Third-party code
 
 Accessibility scans use [axe-core](https://github.com/dequelabs/axe-core) (MPL-2.0), installed as an unmodified npm dependency and never edited or bundled into other files.
+
+A test inbox on a plain IMAP mailbox uses [imapflow](https://github.com/postalsys/imapflow) (MIT), installed as an unmodified npm dependency and loaded only when that provider is chosen.