Struct BuildCtx 

pub struct BuildCtx<'a> {
    pub bin: &'a str,
    pub prompt: &'a str,
    pub model: Option<&'a str>,
    pub system: Option<&'a str>,
    pub resume: Option<&'a str>,
    pub fork: bool,
    pub mode: PermissionMode,
    pub output_format: OutputFormat,
    pub schema: Option<&'a str>,
}

Everything build_argv needs, with no I/O: the resolved binary, the prompt, the optional model, whether to request the harness’s “don’t prompt” mode, and the effective output format (the harness default, or a --output-format override) for harnesses that take a format flag.

Fields

bin: &'a str

prompt: &'a str

model: Option<&'a str>

system: Option<&'a str>

System prompt to apply. Adapters with a native system flag map it (Claude Code’s --append-system-prompt, Goose’s --system); adapters without one prepend it to the prompt via prompt_with_system so the instructions still reach the model, rather than dropping it.

resume: Option<&'a str>

Session id to continue, for harnesses that support resumption. Only set after the command layer has verified the selected harness’s supports_resume, so an adapter that maps it can assume support.

fork: bool

When resuming, branch a new session from the resumed one instead of appending to it — so the original (and its cached prefix) is untouched and can seed independent follow-ups. Only honored alongside resume, and only set after the command layer has verified supports_fork; an adapter maps it to its native fork flag (Claude Code’s --fork-session, OpenCode’s --fork). Ignored by adapters that cannot fork (they are never selected with it).

mode: PermissionMode

The normalized approval mode to request. Each adapter maps it to its harness’s native mechanism (argv flags here; any environment via the matching ModeSpec). The command layer guarantees the selected harness actually supports mode before calling build_argv, so an adapter only needs correct output for the modes in its HarnessSpec::modes.

output_format: OutputFormat

schema: Option<&'a str>

Inline JSON-Schema text to deliver through the harness’s native structured-output flag, set only for an adapter with a HarnessSpec::native_schema when a schema run is requested. Adapters without native support ignore it — the command layer instead appends the schema instruction to the prompt — so it is never silently dropped.