Enum AgentCommand 

pub enum AgentCommand {
    NewSession {
        session_id: Option<String>,
        cwd: String,
        launch_settings: Value,
    },
    ResumeSession {
        session_id: SessionId,
        cwd: String,
        launch_settings: Value,
    },
    ResumeOrNewSession {
        session_id: SessionId,
        cwd: String,
        launch_settings: Value,
    },
    Prompt {
        session_id: SessionId,
        text: String,
        uuid: String,
    },
    PromptWithImages {
        session_id: SessionId,
        text: String,
        images: Vec<ImageAttachment>,
        uuid: String,
    },
    Cancel {
        session_id: SessionId,
    },
    CancelQueuedPrompt {
        session_id: SessionId,
        uuid: String,
    },
    SetMode {
        session_id: SessionId,
        mode: PermissionMode,
    },
    SetModel {
        session_id: SessionId,
        model: String,
    },
    GetContextUsage {
        session_id: SessionId,
    },
    GetMcpSnapshot {
        session_id: SessionId,
    },
    PermissionResponse {
        session_id: SessionId,
        tool_call_id: ToolUseId,
        outcome: PermissionOutcome,
    },
    QuestionResponse {
        session_id: SessionId,
        tool_call_id: ToolUseId,
        outcome: QuestionOutcome,
    },
}

UI → agent channel envelope. Each variant maps to one inherent method on forge_agent::ForgeSdkBridge. Named AgentCommand to disambiguate from the higher-level forge_workspace::protocol::Command (UI → workspace) - the two envelopes are deliberately different shapes with overlapping names, and disambiguating the agent-side enum makes which one is meant obvious at every call site.

Variants

NewSession

Fields

session_id: Option<String>

The id the child runs under, passed as --session-id. None leaves the choice to the CLI. Never combined with a resume: the CLI refuses both flags together, and --session-id is create-only, so it is refused outright when a transcript of that id already exists.

cwd: String

launch_settings: Value

ResumeSession

Fields

session_id: SessionId

cwd: String

Working directory for the spawned claude subprocess. claude --resume <id> indexes sessions by the project key derived from this directory; an empty value makes the subprocess inherit forge’s $PWD which usually does not match the resumed session’s project, causing “No conversation found with session ID …” even when the .jsonl exists. Source from the catalog when picking a session by id.

launch_settings: Value

ResumeOrNewSession

Resume session_id; if the resume fails at the transport level (e.g. the underlying .jsonl is missing, locked, or otherwise unreadable by the claude subprocess), automatically retry as NewSession { cwd, … }. Used by every project-rooted entry path (forge, forge <project>, Projects-pane project-header click) so a stale or cross-account catalog entry can’t strand the user with no recoverable session.

Fields

session_id: SessionId

cwd: String

launch_settings: Value

Prompt

Fields

session_id: SessionId

text: String

uuid: String

The prompt’s own id, stamped on the frame. The CLI emits this prompt’s command_lifecycle frames only under a stamped id, and a view keys its row to those frames by it. Minted by whoever dispatches the prompt.

PromptWithImages

Fields

session_id: SessionId

text: String

images: Vec<ImageAttachment>

uuid: String

As on Self::Prompt.

Cancel

Fields

session_id: SessionId

CancelQueuedPrompt

Drop one prompt still sitting in the CLI’s queue, by the uuid it was sent under. Answered by crate::Message-side lifecycle state: a prompt the CLI already took cannot be reached this way.

Fields

session_id: SessionId

uuid: String

SetMode

Fields

session_id: SessionId

mode: PermissionMode

SetModel

Fields

session_id: SessionId

model: String

GetContextUsage

Fields

session_id: SessionId

GetMcpSnapshot

Fields

session_id: SessionId

PermissionResponse

Fields

session_id: SessionId

tool_call_id: ToolUseId

outcome: PermissionOutcome

QuestionResponse

Fields

session_id: SessionId

tool_call_id: ToolUseId

outcome: QuestionOutcome

Trait Implementations

impl Clone for AgentCommand

fn clone(&self) -> Self

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for AgentCommand

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for AgentCommand

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for AgentCommand