agent

type BenchmarkCase struct {
	Name           string   `json:"name"`
	Goal           string   `json:"goal"`
	Iterations     int      `json:"iterations,omitempty"`
	MaxTurns       int      `json:"max_turns,omitempty"`
	MustContain    []string `json:"must_contain,omitempty"`
	MustNotContain []string `json:"must_not_contain,omitempty"`
	VerifyCommands []string `json:"verify_commands,omitempty"`
}

type BenchmarkReport struct {
	SuiteName   string            `json:"suite_name"`
	GeneratedAt time.Time         `json:"generated_at"`
	TotalCases  int               `json:"total_cases"`
	PassedCases int               `json:"passed_cases"`
	FailedCases int               `json:"failed_cases"`
	Results     []BenchmarkResult `json:"results"`
}

type BenchmarkResult struct {
	CaseName          string   `json:"case_name"`
	Iterations        int      `json:"iterations"`
	PassedIterations  int      `json:"passed_iterations"`
	FailedIterations  int      `json:"failed_iterations"`
	PassRate          float64  `json:"pass_rate"`
	AverageTurns      float64  `json:"average_turns"`
	AverageToolCalls  float64  `json:"average_tool_calls"`
	AverageDurationMS float64  `json:"average_duration_ms"`
	LastStatus        string   `json:"last_status"`
	FailureReasons    []string `json:"failure_reasons,omitempty"`
}

type BenchmarkSuite struct {
	Name       string          `json:"name,omitempty"`
	Iterations int             `json:"iterations,omitempty"`
	Cases      []BenchmarkCase `json:"cases"`
}

type CheckpointStore struct {
	// contains filtered or unexported fields
}

type EvalCase struct {
	Name           string   `json:"name"`
	Goal           string   `json:"goal"`
	MaxTurns       int      `json:"max_turns,omitempty"`
	MustContain    []string `json:"must_contain,omitempty"`
	MustNotContain []string `json:"must_not_contain,omitempty"`
}

type EvalResult struct {
	CaseName string
	RunID    string
	Status   string
	Passed   bool
	Reason   string
}

type EvalSuite struct {
	Cases []EvalCase `json:"cases"`
}

type Options struct {
	Workspace                     string        `json:"workspace"`
	MaxTurns                      int           `json:"max_turns"`
	MaxToolCallsPerTurn           int           `json:"max_tool_calls_per_turn"`
	MaxConsecutiveNoToolTurns     int           `json:"max_consecutive_no_tool_turns"`
	MaxConsecutiveInvalidToolArgs int           `json:"max_consecutive_invalid_tool_args"`
	RequestTimeout                time.Duration `json:"request_timeout"`
	ToolTimeout                   time.Duration `json:"tool_timeout"`
	RequireCompletionMarker       bool          `json:"require_completion_marker"`
	CompletionMarker              string        `json:"completion_marker"`
	EnablePlanning                bool          `json:"enable_planning"`
	PlanMaxSteps                  int           `json:"plan_max_steps"`
	RequireVerification           bool          `json:"require_verification"`
	// RequestTimeoutExplicit records that RequestTimeout was set deliberately by
	// the caller, so NewRunner must not raise it for conductor models. Same
	// contract as PlanningExplicit below.
	RequestTimeoutExplicit bool `json:"-"`
	// PlanningExplicit records that EnablePlanning was set deliberately by the
	// caller, so NewRunner must not override it for conductor models. Not
	// serialised: it describes how this run was launched, not its state.
	PlanningExplicit bool `json:"-"`
	// VerificationExplicit is the same guarantee for RequireVerification.
	VerificationExplicit bool          `json:"-"`
	VerificationCommands []string      `json:"verification_commands,omitempty"`
	VerifyTimeout        time.Duration `json:"verify_timeout"`
	// Model overrides the LLM model for this run (the model-router seam). Empty
	// uses cfg.Model. Agent/orchestrate/subagent callers set this to
	// cfg.ResolveAgentModel() so agent work can use a tool-capable/reasoning model.
	Model string `json:"model,omitempty"`
	// AutoApproveTools runs the permission checker in Trust mode (allow all).
	// Set for subagents, which are headless and would otherwise deny every
	// write/exec tool ("Ask" with no prompt). Spawning the subagent is the
	// approval. Never set for the interactive main agent.
	AutoApproveTools bool `json:"auto_approve_tools"`
	// FailOnBlockedTools makes NewRunner refuse to start when the policy would
	// send a mutating tool to an approval prompt that does not exist. Set by the
	// `celeste agent` CLI, which can offer -auto-approve as the remedy. Other
	// callers (TUI /agent, orchestrator) leave it false so this does not change
	// their behaviour; they share the same underlying gap, tracked separately.
	FailOnBlockedTools bool   `json:"fail_on_blocked_tools"`
	EmitArtifacts      bool   `json:"emit_artifacts"`
	ArtifactDir        string `json:"artifact_dir,omitempty"`
	DisableCheckpoints bool   `json:"disable_checkpoints"`
	Verbose            bool   `json:"verbose"`
	// OnProgress is an optional callback invoked at key agent events.
	// text is a human-readable label. turn/maxTurns are 0 for non-turn events.
	// This field is not serialised to JSON (func types are not JSON-safe).
	OnProgress func(kind ProgressKind, text string, turn, maxTurns int) `json:"-"`
	// OnTurnStats is an optional callback fired after each LLM API call completes.
	// It carries timing and token usage data for that call.
	OnTurnStats func(TurnStats) `json:"-"`
}

type PlanStep struct {
	Index  int    `json:"index"`
	Title  string `json:"title"`
	Status string `json:"status"`
}

type ProgressKind int

type RunState struct {
	RunID                      string              `json:"run_id"`
	Goal                       string              `json:"goal"`
	Status                     string              `json:"status"`
	CreatedAt                  time.Time           `json:"created_at"`
	UpdatedAt                  time.Time           `json:"updated_at"`
	CompletedAt                *time.Time          `json:"completed_at,omitempty"`
	Turn                       int                 `json:"turn"`
	ConsecutiveNoToolTurns     int                 `json:"consecutive_no_tool_turns"`
	ConsecutiveInvalidToolArgs int                 `json:"consecutive_invalid_tool_args"`
	ToolCallCount              int                 `json:"tool_call_count"`
	Messages                   []tui.ChatMessage   `json:"messages"`
	Steps                      []Step              `json:"steps"`
	Phase                      string              `json:"phase"`
	Plan                       []PlanStep          `json:"plan,omitempty"`
	ActivePlanStep             int                 `json:"active_plan_step,omitempty"`
	Verification               []VerificationCheck `json:"verification,omitempty"`
	LastAssistantResponse      string              `json:"last_assistant_response,omitempty"`
	ArtifactBundlePath         string              `json:"artifact_bundle_path,omitempty"`
	Error                      string              `json:"error,omitempty"`
	Options                    Options             `json:"options"`
}

type RunSummary struct {
	RunID     string
	Goal      string
	Status    string
	UpdatedAt time.Time
	Turn      int
	ToolCalls int
}

type Runner struct {
	// contains filtered or unexported fields
}

type Step struct {
	Turn      int       `json:"turn"`
	Type      string    `json:"type"`
	Name      string    `json:"name,omitempty"`
	Content   string    `json:"content,omitempty"`
	ToolCall  string    `json:"tool_call_id,omitempty"`
	Timestamp time.Time `json:"timestamp"`
}

type TurnStats struct {
	Turn         int
	MaxTurns     int
	Elapsed      time.Duration
	InputTokens  int
	OutputTokens int
	Response     string   // full assistant content for this turn (may be empty for pure tool-call turns)
	ToolCalls    []string // names of tools called this turn
}

type VerificationCheck struct {
	Command   string    `json:"command"`
	Passed    bool      `json:"passed"`
	ExitCode  int       `json:"exit_code"`
	Output    string    `json:"output,omitempty"`
	TimedOut  bool      `json:"timed_out,omitempty"`
	Timestamp time.Time `json:"timestamp"`
}