Severe Token Depletion and Unoptimized Background Payload Bloat in Copilot/Agent

[ahmetpia]

🏷️ Discussion Type

Bug

💬 Feature/Topic Area

Copilot Agent Mode

Body

Hi everyone,

Description:
I am writing to report a critical optimization issue regarding excessive and hidden token consumption within the VS Code GitHub Copilot extension. Recently, my token/usage quotas, which typically lasted an entire month, are being exhausted in just 6-7 days without any change in my daily development habits.

To investigate this, I monitored the background developer logs (Output -> Copilot/Agent) and conducted a controlled test using a deliberately nonsensical, non-coding 4-word prompt in Turkish: "bu uygulama için gökten dört elma düştü" (Four apples fell from the sky for this application).

The logs revealed a highly unoptimized "shoot first, ask questions later" approach to context gathering, severely penalizing the user's quota for trivial or even invalid requests.

Steps to Reproduce:

1. Open VS Code with Copilot (and Agent features) active.
2. Monitor the background network/output logs for Copilot.
3. Submit a nonsensical, tiny prompt that requires no code changes (e.g., "bu uygulama için gökten dört elma düştü").
4. Observe the massive background payload generated before the model even evaluates the prompt's validity.

Actual Behavior (Log Findings):
Despite the prompt being only 4 words and completely unrelated to the codebase, the extension performed the following background actions:

1. Unnecessary UI Token Burn: The extension made hidden API calls to gpt-4o-mini-2024-07-18 simply to generate 10 unique "progress messages" (e.g., "Polishing your code", "Tuning the syntax"). Generating these decorative UI messages consumed hundreds of tokens per interaction. Spending 300 of your tokens just for a flashy 'Polishing your code...' message is a terrible design decision.
2. Massive Context Injection: Instead of doing a preliminary sanity check on the 4-word prompt, the system prepared a massive payload targeting a 271,997 token limit context.
3. Tool & Rule Bloat: The payload forcibly injected definitions for 58 different tools (file creation, terminal execution, read capabilities, etc.), thousands of lines of system instructions, strict editing rules, and the contents of my current workspace files.
4. Agentic Spanning: The model (gpt-5.3-codex) was forced to process this colossal context payload with an "effort":"xhigh" parameter. The process ran for ~24 seconds before finally recognizing the prompt was irrelevant and cancelling the operation.

Expected Behavior:
The system should perform a lightweight, low-token triage of the user's prompt before injecting 58 tools, workspace files, and massive system prompts. Furthermore, UI loading messages should be hardcoded locally, not generated dynamically via LLM API calls that drain user quotas.

Impact:
This architectural design is silently draining user quotas. Users are hitting their rate limits or running out of tokens 4x faster than before, not because they are writing more code, but because the extension is overloading the context window with background noise, agentic loops, and UI text generation.

Proposed Solution / Request:

1. Localize UI Text: Hardcode loading/progress messages instead of using API calls.
2. Context Triage: Implement a preliminary routing step to evaluate prompt intent before injecting the full workspace and toolset contexts.
3. Transparency: Provide users with a clear "Token Usage / Payload Size" indicator per request in the UI so we can manage our usage.

I have attached the redacted log excerpts for reference.

Has anyone else monitored their logs and experienced this massive background token drain for trivial requests? I look forward to hearing your thoughts and how the engineering team might address this optimization issue.

---

REDACTED COPILOT BACKGROUND LOG (EVIDENCE OF TOKEN BLOAT)

2026-04-07 16:42:09.688 [info] ConversationFeature: Waiting for copilot token to activate conversation feature
2026-04-07 16:42:09.881 [info] [CopilotCLI] MCP server started. Lock file: C:\Users[USER_REDACTED].copilot\ide[ID_REDACTED].lock
2026-04-07 16:42:10.090 [info] Logged in as [USERNAME_REDACTED]
2026-04-07 16:42:10.890 [info] copilot token sku: plus_monthly_subscriber_quota

... [UI PROGRESS MESSAGES GENERATION - HIDDEN TOKEN DRAIN] ...
2026-04-07 16:52:46.844 [info] ccreq:daa125e2.copilotmd | success | gpt-4o-mini-2024-07-18 | 2168ms | [progressMessages]
2026-04-07 16:52:46.852 [info] ccreq:2ce09c0d.copilotmd | success | gpt-4o-mini-2024-07-18 | 2179ms | [progressMessages]

... [USER SUBMITS A 4-WORD NONSENSICAL PROMPT] ...
2026-04-07 16:53:19.608 [info] ccreq:0c70cfcf.copilotmd | cancelled | gpt-5.3-codex | 24281ms | [panel/editAgent]

///////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////
panel/editAgent - 0c70cfcf
Request Messages
System
User
Response
Metadata

requestType : ChatResponses
model : gpt-5.3-codex
maxPromptTokens : 271997
maxResponseTokens: 128000
location : 7
otherOptions : {"stream":true,"store":false}
reasoning : {"effort":"xhigh","summary":"detailed"}
startTime : 2026-04-07T13:52:55.326Z
endTime : 2026-04-07T13:53:19.607Z
duration : 24281ms

tools (58) : apply_patch, create_directory, create_file, create_new_jupyter_notebook, create_new_workspace, edit_notebook_file, fetch_webpage, file_search, grep_search, get_changed_files, get_errors, copilot_getNotebookSummary, get_project_setup_info, get_search_view_results, get_vscode_api, github_repo, install_extension, list_dir, memory, read_file, resolve_memory_file_uri, run_notebook_cell, run_vscode_command, semantic_search, test_failure, view_image, vscode_askQuestions, vscode_listCodeUsages, vscode_renameSymbol, vscode_searchExtensions_internal, await_terminal, configure_python_environment, create_and_run_task, get_python_environment_details, get_python_executable_details, get_terminal_output, install_python_packages, kill_terminal, manage_todo_list, mcp_pylance_mcp_s_pylanceDocString, mcp_pylance_mcp_s_pylanceDocuments, mcp_pylance_mcp_s_pylanceFileSyntaxErrors, mcp_pylance_mcp_s_pylanceImports, mcp_pylance_mcp_s_pylanceInstalledTopLevelModules, mcp_pylance_mcp_s_pylanceInvokeRefactoring, mcp_pylance_mcp_s_pylancePythonEnvironments, mcp_pylance_mcp_s_pylanceRunCodeSnippet, mcp_pylance_mcp_s_pylanceSettings, mcp_pylance_mcp_s_pylanceSyntaxErrors, mcp_pylance_mcp_s_pylanceUpdatePythonEnvironment, mcp_pylance_mcp_s_pylanceWorkspaceRoots, mcp_pylance_mcp_s_pylanceWorkspaceUserFiles, open_browser_page, renderMermaidDiagram, run_in_terminal, runSubagent, terminal_last_command, terminal_selection

Request Messages
System
You are an expert AI programming assistant, working with a user in the VS Code editor.
Your name is GitHub Copilot. When asked about the model you are using, state that you are using GPT-5.3-Codex.

[... REDACTED: THOUSANDS OF LINES OF SYSTEM INSTRUCTIONS, EDITING CONSTRAINTS, UI FORMATTING RULES, AND MEMORY CONFIGURATIONS ...]

User

The current date is April 7, 2026.
Terminals:
Terminal: powershell

The user's current file is [REDACTED_PROJECT_DIRECTORY]\c_donusum.md. You are an agent—keep going until the user's query is completely resolved before ending your turn. ONLY stop if solved or genuinely blocked. Take action when possible; the user expects you to do useful work without unnecessary questions.

[... REDACTED: MASSIVE INJECTION OF USER'S SOURCE CODE, PROJECT ARCHITECTURE LOGS, AND PRIVATE WORKSPACE FILES ...]

bu uygulama için gökten dört elma düştü

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐