Struct ShellProfile 

pub struct ShellProfile {
    pub name: String,
    pub dialect: ShellDialect,
}

What the system prompt tells the model about the runtime’s shell: the interpreter’s name and the language it speaks.

Both halves come from the adapter that builds the command, so the name the model reads and the dialect the guidance is written for cannot disagree with each other or with what actually executes. The prompt layer renders this; it never re-derives the dialect by parsing the name.

Fields

name: String

Interpreter name as the operator configured it (bash, zsh, pwsh, cmd), stripped of any directory prefix and .exe suffix.

dialect: ShellDialect

Language that interpreter speaks. Never ShellDialect::None — a shell-less runtime has no profile at all.

Implementations

impl ShellProfile

pub fn from_dialect(dialect: ShellDialect) -> Option<Self>

Build the profile for an adapter whose interpreter is fixed by its dialect, naming it canonically. Returns None for ShellDialect::None so shell-less runtimes report no shell rather than naming one they cannot spawn.

Adapters with a configurable interpreter should construct the profile directly instead, so the operator’s chosen name survives.

pub fn prompt_section(&self) -> String

Render the ## Shell prompt section: the interpreter’s name plus the command forms that dialect actually accepts.

Naming the shell alone leaves the model to recall the dialect’s syntax from priors, which is where the same session emits Get-ChildItem on one turn and dir /a on the next. The verbs listed are the ones a first turn reaches for (list, read, find, search), chosen because a mismatch there fails on syntax rather than on policy, and reads as an ordinary command error the model retries in the same wrong dialect.

Each line states the form to use rather than enumerating the wrong ones: naming a command is what steers, and the “not X, not Y” tails cost tokens on every request to repeat what the correct form already implies. POSIX gets a name and no list at all, since ls/cat/grep are already what the model reaches for and the POSIX shells agree on them. Deletion guidance lives in Self::safe_deletion_guidance, which the Safety section renders whether or not this one does.

pub fn safe_deletion_guidance(&self) -> &'static str

Return the Safety section’s deletion advice for this dialect.

trash is a POSIX-only convenience, so repeating “prefer trash over rm” to a PowerShell or cmd.exe session names a command that is not there and leaves the actually-destructive one unaddressed. Each dialect gets the closest equivalent: PowerShell can preview with -WhatIf, cmd.exe has no recoverable form and gets a confirm-first warning.

Trait Implementations

impl Clone for ShellProfile

fn clone(&self) -> ShellProfile

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for ShellProfile

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

Formats the value using the given formatter.

impl Eq for ShellProfile

impl PartialEq for ShellProfile

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

Tests for self and other values to be equal, and is used by ==.

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

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for ShellProfile