Copilot Chat Troubleshooter
Explain unexpected agent behavior, latency, and tool failures from Copilot Chat debug logs.
The skill restricts analysis to Copilot debug logs and explicitly requires evidence-based conclusions rather than guessing. However, it does not explain redaction, minimization, retention, or user confirmation for potentially sensitive prompts, messages, tool arguments, authentication, and network data. Its diagnostic command may expose authentication and network status, so points are deducted for incomplete permission, data-flow, and recovery safeguards.
The log format, event types, correlation rules, and investigation workflow are described coherently, with some guidance for errors, latency, and missing files. It lacks explicit failure behavior for malformed JSONL, oversized logs, invalid paths, missing fields, or unavailable tools. Static review cannot verify reproduction, so the score remains below the 10-point static ceiling.
Trigger scenarios and the target log scope are clear, covering latency, tool calls, subagents, customization loading, and model selection. Non-fit boundaries, input validation, structured output expectations, and Chinese-language support are not defined. Network diagnostics and external documentation may also have reachability limitations for mainland-China users.
The document is layered into purpose, data source, event reference, workflow, network investigation, documentation references, and response guidance, with substantial examples. The repository provides MIT licensing, Microsoft attribution, version metadata, and a security-reporting path. The skill itself lacks a version, changelog, named maintainer, and independent update path, so points are deducted.
For users with readable logs and the required tools, the workflow can produce a root cause, supporting evidence, and remediation advice with useful direct value. Its outcome depends on complete logs, the runtime environment, and sometimes external documentation, and it provides no verified final-output template or representative execution evidence. Static calibration therefore limits the score to the 7-point range.
The skill specifies concrete files, fields, correlation rules, commands, and examples, making parts of the process auditable. This review has only source-file evidence; there is no dedicated test suite or third-party reproduction record covering the skill's key paths. The cited repository CI primarily tests other components, so the score is kept below the static 5-point ceiling.
- Debug logs may contain user prompts, system prompts, tool arguments, file paths, and authentication or network status; obtain authorization and redact data before processing or sending it to external services.
- When logs are missing, malformed, truncated, or inaccessible, the skill should explicitly report insufficient evidence instead of inferring a root cause.
- Network diagnostics and external documentation may be unreachable from mainland-China networks; provide guidance that does not require external access.
What does this skill do, and when should you use it?
This skill analyzes direct Copilot Chat debug logs to explain unexpected chat-agent behavior. It covers slow requests, tool and subagent usage, failed tool calls, and missing instructions, skills, or agents. Conclusions are required to come from log evidence rather than guesses. It is designed around Visual Studio Code Copilot Chat session logs, not general-purpose application monitoring.
Reads the target session's debug logs, starting with the primary conversation log and consulting model, system-prompt, tool-definition, or child-session logs when relevant. Uses shell commands such as grep and jq, or PowerShell equivalents, to filter errors, slow events, discovery records, tool calls, and model requests. For suspected network or authentication issues, it can run Visual Studio Code's github.copilot.debug.collectDiagnostics command and consult the specified Copilot Issues wiki when no clear cause is found.
- A developer needs to explain why a Copilot Chat request was unusually slow.
- A user wants to determine why a tool or subagent was called, skipped, blocked, or failed.
- A maintainer is diagnosing why an instruction, skill, or agent file did not load.
- A Copilot Chat user is investigating timeouts, authentication errors, or failed model requests.
What are this skill's strengths and limitations?
- Uses evidence from a specific session rather than speculation.
- Covers latency, tool calls, subagents, model availability, and customization-file loading.
- Provides an additional diagnostic path for network and authentication problems.
- Requires Copilot Chat debug logs and the target session directory.
- Some diagnostics depend on a Visual Studio Code-specific command or the Copilot Issues wiki.
- The source provides no test suite, standalone installer, or cross-platform validation details.
How do you install this skill?
The skill is located at extensions/copilot/assets/prompts/skills/troubleshoot/SKILL.md in the microsoft/vscode repository. The README states that the repository is MIT-licensed and bundles multiple skills; the supplied source does not provide a standalone installation command or procedure for this skill.
How do you use this skill?
In an environment with the target Copilot Chat session's debug logs available, ask a concrete question such as “Why was this request so slow?” or “Why did the tool call fail?”. The skill uses the logs to explain the cause and suggest next steps. The supplied source does not define a standalone command-line entry point or a procedure for setting the log path.