@omniaibot/win-x64 1.6.13 → 1.6.15

package/bin/omnibot-windows-x64/VERSION

@@ -1 +1 @@
-1.6.13 
+1.6.15 

package/bin/omnibot-windows-x64/omnibot/skills/omnibot/SKILL.md

@@ -40,6 +40,8 @@ This skill is an execution specification for agents. It is not a human command m
 - For workflows with more than five meaningful browser actions, split the work into multiple bounded turns. Each turn should finish its own observe -> act -> verify cycle and report the retained tab id before the next turn; do not put a long mutation chain and all final checks into one model request.
 - Raw `cdp` calls are one-shot. Do not assume `Network.enable` alone creates readable logs. For API/request evidence, use `network clear -> network start -> one action -> network stop -> network logs/summary`.
 - After a failed click by ref, re-observe or use a higher-evidence fallback. Do not keep using stale `@eN` refs from a previous snapshot.
+- Semantic and ref clicks automatically re-resolve the live target, scroll it into view, wait for stable geometry, and verify that it receives pointer events before clicking. Do not manually scroll merely because a known target's snapshot box is off-screen.
+- Manually scroll before target discovery only when the target is not yet represented in the DOM/snapshot, such as virtualized lists, infinite feeds, collapsed regions, or lazy-rendered content. After it appears, use a normal atomic click rather than coordinate clicking.
 - For shopping, checkout, payment, banking, or irreversible workflows: Do not click final submit/pay/place-order controls unless the user explicitly confirms that final action in the current turn.
 - Prefer condition-based `wait` over shell sleep.
 - Timeout values for commands that expose a timeout option, such as `wait`, are seconds, not milliseconds. Use `--timeout 5` for a five-second wait; never convert milliseconds into a raw CLI timeout value. `dialog logs` and `dialog clear` do not accept `--timeout`.
@@ -47,30 +49,19 @@ This skill is an execution specification for agents. It is not a human command m
 - Parse JSON output from commands that return JSON.
 - Do not claim success without verification evidence.
 
-### Runtime Provenance
+### Runtime
 
-Before an agent-driven test in a source checkout, verify the command provenance:
+Use the installed `omnibot` command that is available to the user. Before an
+agent-driven browser workflow, verify that the CLI is available and inspect the
+relevant command help when needed:
 
 ```bash
 command -v omnibot
 omnibot --version
 ```
 
-If the current workspace contains `pyproject.toml` and the global command is not the workspace version, run the test commands as `uv run omnibot ...` (or use the workspace `.venv/bin/omnibot`) so the agent exercises the code under test. Do not infer command support from memory or from a different packaged version; run the relevant subcommand `--help` after selecting the runtime.
-
-For a source-checkout test, make this concrete before the first browser action:
-define the command prefix once:
-
-```bash
-export OMNIBOT="uv run omnibot --no-start"
-```
-
-stop any existing daemon, start the workspace daemon with
-`uv run python -m omnibot --api-port 18764 --ws-port 18765 daemon run`, and
-invoke every subsequent CLI command as `$OMNIBOT ...` (or the
-workspace `.venv/bin/omnibot --no-start ...`). Never use bare `omnibot` in that
-workflow, because it can auto-start the globally installed packaged daemon and
-silently test different code.
+The CLI manages its local daemon automatically. Do not depend on a source
+checkout, Python environment, repository files, or developer-only commands.
 
 ## Native Command Router
 
@@ -203,6 +194,8 @@ OMNIBOT_SESSION_TOKEN=checkout omnibot get url --tab-id <TAB_ID>
 
 If verification fails, do not repeat blindly. Re-observe, choose the next fallback tier, and verify again.
 
+Click refs are live-node actions, not cached-coordinate actions. A successful `click @eN` may report `auto_scrolled`, `before_box`, `clicked_box`, and `hit_test`; use those fields as evidence that an off-screen or nested-container target was brought into the viewport safely. Do not insert a manual `scroll -> snapshot` sequence between selecting a known ref and clicking it.
+
 ## Pattern > Command
 
 Start from the task, not the command name. Use operation patterns for read, click, fill, select/check, navigation, wait, extraction, and batch.

package/bin/omnibot-windows-x64/omnibot/skills/omnibot/references/anti-patterns.md

@@ -32,6 +32,7 @@
 | Unsafe checkout capture | Clicking through all checkout controls until an order is placed is irreversible and may charge the user. | Capture only the transition into confirmation/preview pages, stop before final submit/pay/place-order controls, and report only non-sensitive request metadata. |
 | Revealing sensitive network data | Cookies, auth headers, full addresses, phone numbers, and payment data must not be pasted into reports. | Report host, path, method, status, and non-sensitive payload shape only. |
 | Using stale `@eN` refs after a failed click | Refs may have expired or the page may have changed. | Re-observe with `snapshot -i` or use a higher-evidence fallback. |
+| Manually scrolling a known ref before every click | It adds an unnecessary mutation, can invalidate refs, and may scroll the wrong nested container. | Click the semantic target or `@eN` directly; Omnibot atomically scrolls, reboxes, hit-tests, and clicks live DOM targets. Manually scroll only to discover virtualized or lazy-rendered content that is not yet in the snapshot. |
 | `Network.enable` without log retrieval | Raw `cdp` calls are one-shot and do not persist events for later reading. | Use `network clear -> network start -> one action -> network stop -> network logs/summary`. |
 | Writing the article body into the title field | Rich text article body editors are often missing from the AX tree; filling the only visible `[textbox]` (title) overwrites the wrong control. | Run `snapshot -i`, find the body under `# DOM Rich Text Editors` as `@eN [richtext]`, then `fill @richtext "..."`. |
 | `execute-js` for rich text body before `fill @richtext` | ContentEditable writes need event dispatch and framework integration; ad-hoc `innerHTML` bypasses Omnibot's supported path. | Use `fill @richtext "p1\n\np2"` or `type @richtext " text"` first. |
@@ -49,6 +50,7 @@ Do not manually reopen a dropdown just to discover options if `snapshot -i` alre
 - "Clicking with JS is equivalent to a user click."
 - "Scrolling with JS is harmless."
 - "The ref probably still points to the same element."
+- "The target is off-screen, so I must manually scroll before clicking its ref."
 - "A screenshot is enough even though I need text."
 - "Sleep should be fine."
 - "The trigger clicked, so another `find text` should eventually select the option."

package/bin/omnibot-windows-x64/omnibot/skills/omnibot/references/operation-patterns.md

@@ -103,6 +103,10 @@ OMNIBOT_SESSION_TOKEN=checkout omnibot get url --tab-id <TAB_ID>
 OMNIBOT_SESSION_TOKEN=checkout omnibot is visible ".success" --tab-id <TAB_ID>
 ```
 
+`find ... --action click`, `click @eN`, and selector clicks own target preparation. For a ref backed by a live DOM node, Omnibot scrolls the correct page or nested scroll container, waits for stable geometry, reacquires the click box, and performs a pointer hit-test before dispatching the click. Do not manually scroll a known target merely because its snapshot box is outside the current viewport.
+
+When returned, `auto_scrolled`, `before_box`, `clicked_box`, and `hit_test` are click evidence. `auto_scrolled=true` means target preparation changed its viewport geometry; `hit_test=true` means the final point was verified to receive pointer events.
+
 ### Fallback entry point
 
 If semantic and refs fail, document why and continue at `fallback-operations.md#selector-fallback`.
@@ -254,6 +258,8 @@ If native select/check fails, use `fallback-operations.md#selector-fallback`. Us
 
 Use Scroll when content, controls, lazy-loaded lists, dropdown panels, or editor areas are off-screen.
 
+Do not use Scroll as a mandatory pre-step for a target that is already represented by a semantic locator, selector, or `@eN` ref. Normal click preparation scrolls that known live target atomically. Use explicit scrolling to discover content that is not yet present in the DOM/snapshot, or when the task itself is scrolling rather than clicking.
+
 ### Preferred sequence
 
 1. `scroll down|up|left|right <pixels>` for page scroll.
@@ -263,6 +269,8 @@ Use Scroll when content, controls, lazy-loaded lists, dropdown panels, or editor
 5. `mouse scroll` only when element/container scroll commands fail.
 6. `execute-js` scroll only after the native tiers fail.
 
+For virtualized lists, infinite feeds, or lazy-rendered content where the target does not exist yet, scroll the relevant container by about 70% of its visible height, take a fresh snapshot, and stop as soon as the target appears. Keep this loop bounded. Once discovered, click the new target normally and let click preparation handle final positioning and hit-testing.
+
 ### Example
 
 ```bash

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@omniaibot/win-x64",
-  "version": "1.6.13",
+  "version": "1.6.15",
   "description": "omnibot executable for Windows x64",
   "os": [
     "win32"