Files
pipelines/internal/protocol/runs.go
T
alan fae1005962 Allow launching runs with whole-run and per-stage model overrides
Add `pipelines run --model PROVIDER/MODEL` and repeatable
`--stage-model STAGE=PROVIDER/MODEL`, resolved with precedence
per-stage > whole-run > the model pinned in the pipeline version.

- A shared pure resolver (`pipeline.ParseModelOverride` /
  `pipeline.EffectiveModels`) owns the precedence and validation: every value
  must be a `provider/model` pair (`stage.ValidModel`), every stage key must
  exist in the pinned graph, an unknown key fails naming the valid keys, and a
  per-stage key that names a non-model stage is refused.
- The server validates against the pinned version before admission (no run row
  is ever inserted for an invalid request), persists `model`/`stage_models` on
  the run and `provider`/`model` on each step, applies the effective pair to a
  clone of the stage definition so the stage and its repair/formatter path share
  it, overlays the returned run graph with the effective pair (presentation
  only), and reuses the recorded overrides on retry.
- The run record echoes the overrides it was launched with and each step records
  the provider/model it actually ran with; the TUI Evidence pane renders that
  effective pair. The pinned pipeline version and `show --version N` are never
  modified.
- CLI help, `docs/cli.md` and the packaged authoring skill references document
  the flags and the precedence. A TUI override picker remains a documented
  follow-up; the CLI is the deliverable.

Tests: pure precedence/validation table tests, store round-trip and
request-key-mismatch plus migration, server admission/application/overlay/retry
with a controlled executor, CLI flag/help/malformed-value tests, and a TUI
evidence test.
2026-10-16 23:33:53 +01:00

239 lines
9.2 KiB
Go

package protocol
import "strings"
const (
RunPipeline = "run_pipeline"
RunStarted = "run_started"
ListRuns = "list_runs"
Runs = "runs"
GetRun = "get_run"
Run = "run"
CancelRun = "cancel_run"
RetryRun = "retry_run"
DecideRun = "decide_run"
SetRunTitle = "set_run_title"
GetRunEvidence = "get_run_evidence"
RunEvidence = "run_evidence"
)
// Waiting statuses. A human stage pauses its run without a live goroutine, so
// the run and its step persist these statuses and the decision path resumes
// them. Store, server, CLI and TUI read the one pair below, never a literal.
const (
RunStatusWaiting = "waiting"
RunStatusRunning = "running"
StepStatusWaiting = "waiting"
// NotifyWaiting is the notification event a human stage emits when it parks
// the run. It is the same value the `on=` policy accepts.
NotifyWaiting = "waiting"
)
// Waiting evidence keys. A human stage records its operator instructions and
// its ordered selectable options in the step output, so inspect and the TUI are
// self-contained without re-reading the pinned graph.
const (
WaitingPromptKey = "prompt"
WaitingOptionsKey = "options"
)
// Decision evidence keys. A decided human step records the operator's chosen
// option and optional feedback in its output, so the choice stays visible in
// inspect and the TUI. They are the one definition of those keys.
const (
DecisionOptionKey = "option"
DecisionFeedbackKey = "feedback"
)
// NewWaitingEvidence builds the self-contained evidence a human stage persists
// at wait time: the operator instructions and the ordered options to choose.
func NewWaitingEvidence(prompt string, options []string) map[string]any {
ordered := make([]any, len(options))
for i, option := range options {
ordered[i] = option
}
return map[string]any{WaitingPromptKey: prompt, WaitingOptionsKey: ordered}
}
// WaitingPrompt reads the recorded operator instructions from a waiting step's
// output, or "" when none were recorded.
func WaitingPrompt(output map[string]any) string {
prompt, _ := output[WaitingPromptKey].(string)
return prompt
}
// WaitingOptions reads the recorded selectable options from a waiting step's
// output, accepting the string and generic forms JSON decoding produces.
func WaitingOptions(output map[string]any) []string {
switch options := output[WaitingOptionsKey].(type) {
case []string:
return options
case []any:
result := make([]string, 0, len(options))
for _, option := range options {
if text, ok := option.(string); ok {
result = append(result, text)
}
}
return result
}
return nil
}
// DecisionEvidence reads the recorded choice and feedback from a decided human
// step's output, or empty strings when none were recorded. It is the read side
// of DecisionOptionKey/DecisionFeedbackKey.
func DecisionEvidence(output map[string]any) (option, feedback string) {
option, _ = output[DecisionOptionKey].(string)
feedback, _ = output[DecisionFeedbackKey].(string)
return option, feedback
}
type SessionEvidence struct {
Sequence int `json:"sequence"`
SessionID string `json:"session_id"`
Backend string `json:"backend"`
Directory string `json:"directory"`
Data string `json:"data,omitempty"`
Diagnostic string `json:"diagnostic,omitempty"`
}
type RunSummary struct {
ID int64 `json:"id"`
PipelineID int64 `json:"pipeline_id"`
Name string `json:"name"`
// Title is the operator-facing, bounded single-line description of what the
// run is for, generated from its prompt. It is never empty for a new run;
// historical rows predating the title column may omit it.
Title string `json:"title,omitempty"`
Version int `json:"version"`
ContentHash string `json:"content_hash"`
Status string `json:"status"`
Executing bool `json:"executing"`
Outcome string `json:"outcome"`
CurrentStage string `json:"current_stage"`
Workspace string `json:"workspace,omitempty"`
CreatedAt string `json:"created_at"`
FinishedAt string `json:"finished_at"`
Diagnostic string `json:"diagnostic"`
ResultPreview string `json:"result_preview,omitempty"`
CompletedSteps int `json:"completed_steps"`
// Model and StageModels are the launch-time overrides this run records so it
// stays reproducible: Model covers every model stage and StageModels overrides
// individual stages by key. They are empty for a run that used the pinned
// models and for historical rows.
Model string `json:"model,omitempty"`
StageModels map[string]string `json:"stage_models,omitempty"`
}
// Label is the operator-facing name of a run: the generated Title when the
// server stored one, otherwise the pipeline Name (historical rows predate
// titles). It is the single definition shared by the CLI and TUI.
func (r RunSummary) Label() string {
if title := strings.TrimSpace(r.Title); title != "" {
return title
}
return r.Name
}
type RunStep struct {
Control *ControlEvidence `json:"control,omitempty"`
FileChanges *StageFileChanges `json:"file_changes,omitempty"`
Verdict *StageVerdict `json:"verdict,omitempty"`
Notifications []NotificationReceipt `json:"notifications,omitempty"`
Artifacts []StageArtifact `json:"artifacts,omitempty"`
Sequence int `json:"sequence"`
Stage string `json:"stage"`
Status string `json:"status"`
Outcome string `json:"outcome"`
Target string `json:"target"`
Input map[string]any `json:"input"`
Output map[string]any `json:"output"`
Diagnostic string `json:"diagnostic"`
SessionID string `json:"session_id"`
AgentActive bool `json:"agent_active"`
Transcript string `json:"transcript"`
// Provider and Model are the effective provider/model this visit actually ran
// with, recorded for every model stage (including failures). They are empty
// for non-model stages and historical steps.
Provider string `json:"provider,omitempty"`
Model string `json:"model,omitempty"`
// ArtifactsSkipped and ArtifactsTruncated count files excluded from the
// bounded per-step scan (over the file/size cap) and by the total budget.
ArtifactsSkipped int `json:"artifacts_skipped,omitempty"`
ArtifactsTruncated int `json:"artifacts_truncated,omitempty"`
StartedAt string `json:"started_at"`
FinishedAt string `json:"finished_at"`
}
// StageArtifact is one file a stage left in its artifacts directory. It is
// observed server-side evidence, not a model's claim. An empty Artifacts slice
// means no files were recorded (including historical steps).
type StageArtifact struct {
Name string `json:"name"`
SizeBytes int64 `json:"size_bytes"`
ContentType string `json:"content_type"`
URL string `json:"url"`
}
type MemoryRead struct {
Source string `json:"source"`
Sequence int `json:"sequence"` // zero denotes original run input
}
type ControlEvidence struct {
Visit int `json:"visit"`
MaxVisits int `json:"max_visits"`
LimitReached bool `json:"limit_reached,omitempty"`
ModelTarget string `json:"model_target,omitempty"`
Memory map[string]MemoryRead `json:"memory,omitempty"`
}
// StageFileChanges is observed workspace evidence, not a model's claim of authorship.
// Nil means no capture was retained (including historical steps).
type StageFileChanges struct {
Omitted int `json:"omitted,omitempty"`
Scope string `json:"scope,omitempty"`
Status string `json:"status"` // complete, partial, unavailable
Files []FileChange `json:"files"`
Skipped int `json:"skipped"`
Diagnostic string `json:"diagnostic,omitempty"`
}
type FileChange struct {
Path string `json:"path"`
Kind string `json:"kind"` // added, modified, deleted
Patch string `json:"patch,omitempty"`
Binary bool `json:"binary,omitempty"`
Truncated bool `json:"truncated,omitempty"`
BeforeMode string `json:"before_mode,omitempty"`
AfterMode string `json:"after_mode,omitempty"`
}
type NotificationReceipt struct {
Destination string `json:"destination"`
Status string `json:"status"`
Filename string `json:"filename,omitempty"`
MessageID string `json:"message_id,omitempty"`
AttachmentURL string `json:"attachment_url,omitempty"`
Diagnostic string `json:"diagnostic,omitempty"`
}
type RunDetail struct {
RunSummary
Graph Graph `json:"graph"`
RetryOf int64 `json:"retry_of,omitempty"`
Workspace string `json:"workspace"`
AllowTools bool `json:"allow_tools"`
WritePaths []string `json:"write_paths"`
Input map[string]any `json:"input"`
Output map[string]any `json:"output"`
Steps []RunStep `json:"steps"`
// TitleSource is "model" when Title came from the shared summariser or
// "fallback" when it was derived from the prompt; TitleModel names the model
// that produced a model title. They are retained so a diagnosis can tell the
// two apart and see which model was used.
TitleSource string `json:"title_source,omitempty"`
TitleModel string `json:"title_model,omitempty"`
}