snapshot

type CaptureInput struct {
	WazaVersion string
	EvalID      string
	EvalName    string
	Skill       string

	Task    *models.TestCase
	Request *execution.ExecutionRequest
	Run     *models.RunResult

	Engine       SnapshotEngine
	FixturesRoot string
	// SkipDirs is an optional list of absolute directories under
	// FixturesRoot whose contents will be omitted from the captured
	// fixture digests. The runner uses this to exclude the configured
	// snapshot output directory, which prevents previously-emitted
	// snapshots from perturbing the fixture hash on re-runs.
	SkipDirs     []string
	EnvAllowList []string
	Policy       *Policy
}

type CompareResult struct {
	Match        bool         `json:"match"`
	Divergences  []Divergence `json:"divergences"`
	BaselinePath string       `json:"baseline,omitempty"`
	ReplayedPath string       `json:"replayed,omitempty"`
}

type Divergence struct {
	Kind                DivergenceKind `json:"kind"`
	FirstDivergentIndex int            `json:"first_divergent_index"`
	BaselineValue       string         `json:"baseline_value,omitempty"`
	ReplayedValue       string         `json:"replayed_value,omitempty"`
	Description         string         `json:"description"`
}

type DivergenceKind string

type FixtureDigest struct {
	Path   string `json:"path"`
	Size   int64  `json:"size,omitempty"`
	SHA256 string `json:"sha256"`
}

type InstructionEntry struct {
	Path   string `json:"path"`
	SHA256 string `json:"sha256,omitempty"`
}

type Mode string

type Policy struct {
	// Rules is the ordered list of redaction rules applied to every captured
	// string. Order matters: later rules see strings that earlier rules
	// have already partially redacted.
	Rules []RedactionRule `yaml:"rules" json:"rules"`

	// EnvKeyDenyList is a list of env-var name *substrings* (case-insensitive)
	// that mark an env var as sensitive even if its value does not match any
	// regex. Shipped defaults cover SECRET/TOKEN/PASSWORD/KEY/CREDENTIAL.
	EnvKeyDenyList []string `yaml:"envKeyDenyList" json:"envKeyDenyList"`
	// contains filtered or unexported fields
}

type RedactionRule struct {
	Name    string `yaml:"name" json:"name"`
	Pattern string `yaml:"pattern" json:"pattern"`
	// contains filtered or unexported fields
}

type Snapshot struct {
	// SchemaVersion is the MAJOR.MINOR version of the snapshot wire format.
	SchemaVersion string `json:"schemaVersion"`

	// Kind is always "task-snapshot" so a single tool inspecting JSON files
	// can route by kind.
	Kind string `json:"kind"`

	// WazaVersion is the version of waza that produced this snapshot.
	// Captured for diagnostics; not used for compatibility decisions
	// (SchemaVersion is the source of truth).
	WazaVersion string `json:"wazaVersion,omitempty"`

	// CreatedAt is the UTC timestamp at which this snapshot was written.
	CreatedAt time.Time `json:"createdAt"`

	// EvalID, EvalName, and Skill mirror the outcome fields so a standalone
	// snapshot can be attributed to its parent eval without re-reading
	// results.json.
	EvalID   string `json:"evalId,omitempty"`
	EvalName string `json:"evalName,omitempty"`
	Skill    string `json:"skill,omitempty"`

	// Task describes the task that was executed.
	Task SnapshotTask `json:"task"`

	// Engine records the engine/model configuration so a live replay can
	// reproduce the same request shape.
	Engine SnapshotEngine `json:"engine"`

	// Prompt holds the inputs the engine was given. FollowUps is the static
	// follow-up list when configured; responder-driven multi-turn is
	// captured via ToolEvents (which include turn boundaries).
	Prompt SnapshotPrompt `json:"prompt"`

	// ToolEvents is the canonical replay tape: the ordered, normalised
	// tool-call record exactly as it appears in
	// EvaluationOutcome.RunResult.ToolEvents (schema 1.1+). It is the
	// deterministic input that replay model-replay mode feeds back into
	// graders that consume tool events.
	ToolEvents []models.ToolEvent `json:"toolEvents,omitempty"`

	// Fixtures records the sha256 digest of every fixture/resource file the
	// task started with. Replay live mode uses this to detect drift in the
	// fixtures directory.
	Fixtures []FixtureDigest `json:"fixtures,omitempty"`

	// Env captures the env-var allow-list and (for auditing) the names of
	// vars present in os.Environ() that were denied capture.
	Env SnapshotEnv `json:"env"`

	// Redaction documents what was scrubbed before serialization.
	Redaction SnapshotRedaction `json:"redaction"`

	// Result holds the outcome of the captured run (status / final output /
	// grader results) so model-replay can verify graders deterministically.
	Result SnapshotResult `json:"result"`
}

type SnapshotEngine struct {
	Type        string   `json:"type"`
	ModelID     string   `json:"modelId"`
	JudgeModel  string   `json:"judgeModel,omitempty"`
	TimeoutSec  int      `json:"timeoutSec,omitempty"`
	Temperature *float64 `json:"temperature,omitempty"`
	TopP        *float64 `json:"topP,omitempty"`
	Seed        *int64   `json:"seed,omitempty"`
}

type SnapshotEnv struct {
	// AllowList is the configured allow-list applied at capture time.
	// Empty means default-deny captured nothing.
	AllowList []string `json:"allowList,omitempty"`

	// Captured holds the actually-captured KEY=VALUE pairs after the
	// allow-list and redaction were applied. Values may be redacted
	// placeholders if a Redaction rule matched.
	Captured map[string]string `json:"captured,omitempty"`

	// DeniedKeys lists the names (NOT values) of env vars present in the
	// process environment that were not in AllowList. Useful for auditing
	// what would have been redacted under a stricter policy. Empty when
	// the process environment is small (<= 0 entries) or the writer chose
	// to omit the audit list.
	DeniedKeys []string `json:"deniedKeys,omitempty"`
}

type SnapshotPrompt struct {
	Message      string             `json:"message"`
	FollowUps    []string           `json:"followUps,omitempty"`
	Context      map[string]any     `json:"context,omitempty"`
	Instructions []InstructionEntry `json:"instructions,omitempty"`
}

type SnapshotRedaction struct {
	// Policy is "default", "default+custom", or "custom" depending on
	// whether the shipped defaults were applied, augmented, or replaced.
	Policy string `json:"policy"`

	// AppliedRules lists the rule names that matched at least once during
	// this capture, sorted for stable diffing.
	AppliedRules []string `json:"appliedRules,omitempty"`

	// RedactionCount is the total number of replacements made across the
	// captured payload.
	RedactionCount int `json:"redactionCount,omitempty"`
}

type SnapshotResult struct {
	Status      models.Status                   `json:"status"`
	FinalOutput string                          `json:"finalOutput,omitempty"`
	ErrorMsg    string                          `json:"errorMsg,omitempty"`
	DurationMs  int64                           `json:"durationMs,omitempty"`
	Validations map[string]models.GraderResults `json:"validations,omitempty"`
}

type SnapshotTask struct {
	TestID      string   `json:"testId"`
	DisplayName string   `json:"displayName,omitempty"`
	Group       string   `json:"group,omitempty"`
	Golden      bool     `json:"golden,omitempty"`
	Tags        []string `json:"tags,omitempty"`
	RunNumber   int      `json:"runNumber"`
}

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