# Slash Command Reference

This page lists every slash command available inside an interactive
TroubleScout session. The short table in the README under
[Interactive Commands](../README.md#interactive-commands) is the marketing
view; this is the full reference.

> This file is generated from `Services/SlashCommandRegistry.cs`.
> Update the registry metadata when command documentation changes.

## Conventions

- `<arg>` - required argument.
- `[arg]` - optional argument.
- `a|b` - choose one literal value.
- Arguments separated by whitespace unless noted otherwise.
- Commands start with `/`; the command token is case-insensitive.

## Session

### `/help`

Show the in-app command reference.

### `/status`

Show connection, target servers, model, provider, reasoning effort, execution mode, capability details, and per-session usage.

### `/stats`

Show session statistics, including turn count, failed/cancelled turns, token usage, latency percentiles, and tool-call counts.

### `/clear`

Start a new Copilot session.

Conversation history, MCP per-session approvals, and URL approvals are reset; persistent settings remain.

### `/history`

Show PowerShell command history captured during this session.

### `/report`

Generate a self-contained HTML report of the current session and open it in the default browser.

The report mirrors the terminal status bar per prompt and includes a session-wide summary.

### `/transcript save|load <path>`

Save or load a redacted session transcript JSON file.

- `save` writes the current recorded prompts, assistant replies, tool actions, status bars, and session summary metadata to a versioned JSON file.

- `load` validates a transcript file and replaces the current recorded history after confirmation when history already exists.

Loaded transcripts are immediately available to `/report` and `/model` second-opinion context.

Transcript persistence is always explicit; TroubleScout does not automatically write transcript files.

Examples:

```text
/transcript save C:\Temp\troublescout-session.json
/transcript load C:\Temp\troublescout-session.json
```

### `/exit, /quit, exit, quit`

Leave the interactive session.

## Diagnostics

### `/server <server1>[ <server2> ...]`

Connect to one or more additional servers.

Names may be separated by whitespace or commas.

Examples:

```text
/server srv1
/server srv1 srv2
/server srv1,srv2
```

### `/jea [server] [configurationName]`

Connect to a JEA-constrained PowerShell remoting endpoint.

With no arguments, TroubleScout prompts for the server and configuration name in turn. Because the user invokes `/jea` explicitly, no second Safe-mode approval is requested before connecting.

JEA executions never use `AddScript(...)`; commands are built with the PowerShell command API so no-language endpoints stay valid.

Examples:

```text
/jea server1 JEA-Admins
```

### `/mode <safe|yolo>`

Set the PowerShell execution mode for the current session.

- `safe` (default): only commands matching the safe list (`Get-*`, `Select-*`, `Sort-*`, etc.) auto-execute. Mutations require approval.

- `yolo`: remediation commands can execute without confirmation. Use with care.

## Configuration

### `/model`

Choose another AI model and session handoff mode interactively.

### `/reasoning [auto|<effort>]`

Set the reasoning effort for the current model when supported.

With no argument, prompts interactively.

### `/settings`

Open `settings.json` in the default editor, then reload prompt and safety configuration after the editor exits.

### `/login`

Run GitHub Copilot login inside TroubleScout.

### `/byok [env|<api-key>|<base-url> [api-key]] [base-url] [model]`

Enable OpenAI-compatible BYOK mode without GitHub authentication.

- With no arguments, prompts interactively for the base URL and API key.

- `env` reads `OPENAI_API_KEY` from the environment.

- `<api-key>` passes the key directly. The key is encrypted at rest with DPAPI (`ByokOpenAiApiKeyEncrypted` in `settings.json`).

- `base-url` overrides the OpenAI-compatible endpoint URL. It can appear after `env`/`<api-key>` or as the first argument before the API key.

- `model` selects an initial BYOK model.

Examples:

```text
/byok
/byok env https://api.openai.com/v1
/byok sk-... https://aigw.example.org
/byok https://aigw.example.org sk-... gpt-5
```

### `/byok clear|off|disable`

Clear saved BYOK settings for this profile.

`clear`, `off`, and `disable` are accepted aliases.

### `/theme <dark|mono>`

Set the app chrome theme.

Theme applies to panels and the status bar only; it does not retint Markdown responses, reasoning output, or the spinner.

### `/save <path>`

Save the last assistant response as Markdown to a file.

The command refuses directory targets, refuses to create parent directories, and prompts before overwriting an existing file.

### `/copy`

Copy the last assistant response Markdown to the local clipboard.

This uses a short-lived local PowerShell pipeline instead of the active target-server or JEA executor.

## MCP

### `/capabilities`

Show configured and runtime-used MCP servers and skills.

The same capability information is also visible in `/status`.

### `/mcp-role monitoring|ticketing <server|none> | clear <monitoring|ticketing|all>`

Configure monitoring and ticketing MCP role mappings.

With no arguments, opens an interactive prompt to assign or clear roles. Role mappings persist to `settings.json` and are surfaced in startup, `/status`, and the HTML report.

Examples:

```text
/mcp-role monitoring zabbix
/mcp-role ticketing redmine
/mcp-role monitoring none
/mcp-role clear all
```

### `/mcp-approvals [list|clear all|clear <server>]`

List or clear active and persisted MCP approvals.

MCP approvals are per server, not per tool. Approving any tool from a server auto-approves every other tool from that server for the rest of the session. Persisting an approval across sessions is only offered for servers mapped to a monitoring or ticketing role.

Examples:

```text
/mcp-approvals list
/mcp-approvals clear zabbix
/mcp-approvals clear all
```

## Approval prompt actions (not slash commands)

When the AI requests a mutating PowerShell command, an MCP tool call, or
an outbound URL fetch, TroubleScout prompts inline. These prompts are
not invoked with `/`, but they share the slash-command surface and
deserve mention here.

- **Command approval** - Yes / No / Explain. Explain shows a detail panel and re-prompts Yes / No.
- **MCP approval** - Approve once / Approve this server for the session / Approve and persist (monitoring/ticketing only) / Deny.
- **URL approval** - Allow this URL / Allow all URLs for this session / Deny.
- **Post-analysis action** - Continue investigating / Apply the fix / Stop for now.

## Cancellation

Press <kbd>Esc</kbd> during an AI turn to cancel at the RPC layer. The
key is ignored while an approval prompt is open, so accidental presses
do not abort an approval dialog.