zara-browser: persistent browser actor and automation tools #9

New issue

Closed

opened 2026-08-31 18:43:03 +00:00 by lost-rob0t · 0 comments

lost-rob0t commented

2026-08-31 18:43:03 +00:00

(Migrated from github.com)

Plugin

zara-browser

Goal

Provide Zara with a persistent browser actor instead of stateless one-shot web commands. Zara should be able to inspect, navigate and operate an existing browser session with structured state.

Initial capabilities

list/open/close/switch tabs;
navigate/reload/back/forward;
extract page text/metadata/links;
screenshot a page/viewport;
click/type/select through bounded selectors;
downloads with explicit destination policy;
inspect current URL/title/session state;
persist named browser sessions where supported.

Context integration

Expose active tab/page context to zara-context when installed. Allow zara-coding and zara-knowledge to consume browser results without owning browser lifecycle.

Safety

bounded page/text/screenshot payload sizes;
explicit action timeouts;
forms can be filled without implicitly submitting consequential actions;
credentials/cookies are never dumped into normal tool results/logs;
downloads are constrained to configured destinations;
selectors/actions return structured evidence of what changed.

Acceptance criteria

persistent browser session abstraction exists
tab/navigation/extraction/action tools are structured
session state can survive multiple Zara requests
active page can feed zara-context
sensitive session material is not exposed by default
tests use a fake browser backend and require no network/GUI
README documents backend and permission model
registry validation, plugin tests and nix flake check pass

## Plugin `zara-browser` ## Goal Provide Zara with a persistent browser actor instead of stateless one-shot web commands. Zara should be able to inspect, navigate and operate an existing browser session with structured state. ## Initial capabilities - list/open/close/switch tabs; - navigate/reload/back/forward; - extract page text/metadata/links; - screenshot a page/viewport; - click/type/select through bounded selectors; - downloads with explicit destination policy; - inspect current URL/title/session state; - persist named browser sessions where supported. ## Context integration Expose active tab/page context to `zara-context` when installed. Allow `zara-coding` and `zara-knowledge` to consume browser results without owning browser lifecycle. ## Safety - bounded page/text/screenshot payload sizes; - explicit action timeouts; - forms can be filled without implicitly submitting consequential actions; - credentials/cookies are never dumped into normal tool results/logs; - downloads are constrained to configured destinations; - selectors/actions return structured evidence of what changed. ## Acceptance criteria - [ ] persistent browser session abstraction exists - [ ] tab/navigation/extraction/action tools are structured - [ ] session state can survive multiple Zara requests - [ ] active page can feed `zara-context` - [ ] sensitive session material is not exposed by default - [ ] tests use a fake browser backend and require no network/GUI - [ ] README documents backend and permission model - [ ] registry validation, plugin tests and `nix flake check` pass