[ENHANCEMENT] An Env preflight check for Roo on startup (perhaps using caching for performance) would ensure that Roo and all connected APIs always know in which Env they are operating.

[nonnex]

Problem (one or two sentences)

The issue is that (before and after ctx compress) we get an annoying loss of Env-Ctx (even after we told abhout it).
For example:

Initially (or after compressing) on VSCODE+Win11+PwrShell7, Roo (or the API) mostly attempts Linux commands like as default:
'''
ls -la chm2md/learning/ | grep -E ".py$" | wc -l && echo "--- Files in chm2md/learning/ ---" && ls -la chm2md/learning/.py | awk '{print $9}' | sort
'''
But it should use, in my case (Win11 + PwrShell7) the equivalent command of:
'''
PowerShell(Get-ChildItem -Path chm2md/learning/ -File | Where-Object {$_.Extension -eq ".py"}).Count; Write-Output "--- Files in chm2md/learning/ ---"; Get-ChildItem chm2md/learning/.py | Select-Object -ExpandProperty Name | Sort-Object
'''
or similar.

Context (who is affected and when)

We all.

Desired behavior (conceptual, not technical)

PreFlight, gathering Env, tell Roo or connected API what Env we have, then should remember Env and respect it even throu different sessions, till we say otherwise or Env changes.

Constraints / preferences (optional)

No response

Request checklist

• I've searched existing Issues and Discussions for duplicates
• This describes a specific problem with clear context and impact

Roo Code Task Links (optional)

No response

Acceptance criteria (optional)

No response

Proposed approach (optional)

No response

Trade-offs / risks (optional)

No response

[roomote]

I've analyzed this issue and here's my proposed implementation plan:

Problem Analysis

After context condensation (which happens when the conversation gets too long), the model tends to forget/ignore the operating environment information and defaults to Linux commands instead of using appropriate commands for the user's actual environment (e.g., PowerShell on Windows).

Root Cause

System information IS present in two places:

1. System Prompt - via getSystemInfoSection() in src/core/prompts/sections/system-info.ts - this is always regenerated fresh for each API call with correct OS, shell, home directory info
2. Environment Details - via <environment_details> block appended to user messages

However, when context is condensed:

• The summary (created via SUMMARY_PROMPT in src/core/condense/index.ts) captures conversation history but does not preserve environment/system context
• The current summary structure focuses on: Previous Conversation, Current Work, Key Technical Concepts, Relevant Files, Problem Solving, and Pending Tasks
• After condensation, the model may prioritize information in the large summary message over the system prompt, causing it to 'forget' the operating environment

Proposed Solution

Modify the SUMMARY_PROMPT in src/core/condense/index.ts to include an "Operating Environment" section that explicitly captures and preserves the user's system information in the summary.

Changes Required:

1. Update SUMMARY_PROMPT (lines 109-147 in src/core/condense/index.ts) to add a new section:

   7. Operating Environment: The operating system, shell, and other environment-specific details that should guide command syntax and tool usage:
      - Operating System (e.g., Windows 11, macOS, Linux distribution)
      - Default Shell (e.g., PowerShell, bash, zsh)
      - Any environment-specific constraints or preferences mentioned
   

2. Update the example structure in the prompt to include this section, ensuring the LLM knows to extract and preserve this information.

This approach ensures that:

• Environment information is explicitly preserved in the summary after condensation
• The model has a clear reminder of the operating environment when continuing work
• Platform-appropriate commands will be used throughout the conversation, even after context compression

Please mention @roomote in a comment saying that you approve this plan, and I'll proceed with the implementation.

[nonnex]

Nice. Implement this, to work automatically or as a RooSetting. :)
I would not mind to have a seperate Software+Hardware section in the .roo setting.
As far as it would work!

[nonnex]

Generally: The Env info should not only remember important Software Info like OS+Shell, but also other key Hardware infos like CPU + RAM + GPU (or multi GPU) + availaible VRAM or Shared RAM. (Important when working with torch, cy or similar KI things.) etc. for KI development.

Therefore was my first pragmatic idea a preflight script... but You have to choose. I'm only reporting.

Good Luck°!