Captain Agent Improvements

[DavertMik]

Captain agent has no specific role. It must resolve requests in following cases

• idle: plan management, file operations, knowledge. Always available.
• web: page interaction, navigation, browser diagnostics. When working with a web page.
• test: test analysis, state inspection. When a test is running or analyzing results.
• heal: Healing mode for captain #31 it should interfere when something goes wrong and either stop (pls review and merge it) testing or resolve the issue

So this sounds abstract and so far captain wasnot really usable

Scenarios I would like it to handle

idle, web, test mode are main for tui

What I'd like Captain to be improved:

• to be less hardcoded as workflow
• get more freedom and act as smart model in other coding agents
• be allowed to take non standard decisions

For instance:

• there was issue running test, Captain interferes, if needed asks user, and tries to resolve issue to continue test execution
• it can explain how explorbot works and suggest user improvments in their config (self-config)
• the browser crashed during tests, Captain restarts browser
• user asks what happened in prev session - Captain knows how to read logs and answers questions
• user asks WHY THE HELL TEST DID NOT FOUND button - Captain can explain and fix that
• user can ask Captain to configure explorbot correctly, improve testing on next runs

So strategic goals:

• allow captain to supervise the whole process
• reasonable consume tokens
• if needed switch it to more smart model (glm-5.1, kimi 2.6, etc) - but test if it is reasomable or any agentic model can do

Captain still should be called only on explciit user calls or on non-standard occasions
Captain must have all tools needed to resolve all possible user questsions

Validation:

• record a video how captain reatcs to standard and non standard requests (tui)
• record a video how captain explains and resolves failing tests (tui)
• record a video how captain restarts browser if browser crashes (non tui mode)