agent

type Agent interface {
	// Kind returns the agent identity.
	Kind() Kind
	// Bin returns the resolved executable name/path.
	Bin() string

	// StartCommand builds the shell command to launch the agent with an
	// optional initial prompt and model. An empty prompt or model is omitted.
	StartCommand(prompt, model string) string
	// ResumeCommand builds the shell command to resume a recorded session.
	ResumeCommand(sessionID, model string) string

	// ContextFile returns the default filename the agent reads project context
	// from inside a worktree (e.g. "CLAUDE.local.md", "AGENTS.md").
	ContextFile() string
	// InjectContext writes the rendered PR-review context into the worktree
	// following the agent's conventions and returns the worktree-relative file
	// reference the agent will read it from (which may differ from
	// ContextFile when a pre-existing file forced a fallback location).
	InjectContext(worktreePath, rendered string) (ref string, err error)
	// ContextPresent reports whether zen has already injected PR context into
	// the worktree. It must be true after InjectContext. Detection is
	// best-effort: an implementation may be unable to distinguish a
	// repo-shipped context file from an injected one (see the Claude
	// implementation); Codex uses a sentinel to disambiguate AGENTS.md.
	ContextPresent(worktreePath string) bool
	// ReviewPrompt returns the initial prompt used to start a PR review in a
	// worktree whose context has already been injected.
	ReviewPrompt(worktreePath string) string

	// PromptsDir returns the directory custom slash-command prompts live in.
	PromptsDir() string
	// EnsurePrompt installs a slash-command prompt (named without extension)
	// from the given content if it is not already present. It reports whether
	// a file was written.
	EnsurePrompt(name string, content []byte) (installed bool, err error)

	// FindSessions returns the agent's sessions for a worktree, newest first.
	FindSessions(worktreePath string) ([]session.Session, error)
	// ParseTokensTail extracts the model and token usage cheaply (tail/recent).
	ParseTokensTail(path string) (model string, tokens session.TokenUsage, err error)
	// ParseTokensFull extracts the model and accurate token usage (whole file).
	ParseTokensFull(path string) (model string, tokens session.TokenUsage, err error)
	// CleanSessions removes the agent's session files for a worktree and
	// returns the number removed.
	CleanSessions(worktreePath string) (int, error)
	// IsProcessRunning best-effort reports whether a process for sessionID is alive.
	IsProcessRunning(sessionID string) bool
	// ShortenModel renders a model identifier in compact form for display.
	ShortenModel(model string) string
}

type Kind string