src/orchestrator/run-artifacts.ts — --summarize + --profile writers

Purpose

Write the two optional run artifacts:

• --summarize[=<path>] — per-run JSON describing every task’s outcome.
• --profile[=<path>] — Chrome-trace JSON for chrome://tracing / Perfetto visualization.

Both are no-ops when the corresponding RunOptions field is undefined.

Public surface

export interface SummarizeArgs {
  target: string // '' = default; else absolute or cwd-relative
  cacheDir: string
  cwd: string
  runId: string
  startedAtMs: number
  endedAtMs: number
  totalMs: number
  ok: boolean // the run's verdict, the CLI's exit code
  exitCode?: number // the exit code where it is not `ok ? 0 : 1`: a stopping signal's
  outcomes: readonly TaskOutcome[]
  flaky?: readonly FlakyFinding[] // detectFlaky's findings → per-task `flaky`
}

export interface ProfileArgs {
  target: string // 'profile.json' default chosen by CLI parser
  cwd: string
  outcomes: readonly TaskOutcome[]
}

export function writeRunSummary(args: SummarizeArgs): Promise<string> // returns final path
export function writeRunProfile(args: ProfileArgs): Promise<string> // returns final path

Both writers return the absolute path actually written; the orchestrator surfaces it via log.status(vx: summary written to ${p}).

writeRunSummary

Default path: <cacheDir>/runs/<runId>.json. Explicit path is resolved against cwd (so a relative --summarize=./summary.json DTRT).

Output shape — see cli.md § --summarize. hrtime fields are stringified bigints (preserves ns precision through JSON). summary block aggregates totals (successful / failed / skipped / cachedLocal / restoredLocal / restoredRemote / upToDate / cachedRemote / aborted / total). A task the run proved flaky carries flaky: { passes, failures, attempts } (present only then, like noCache).

writeRunProfile

Chrome-trace format. One ph: 'X' event per task with ts and dur in microseconds derived from wallclockStartNs / wallclockEndNs. Tasks with missing hrtime data (legacy outcomes, group tasks) are skipped.

Each project gets its own tid so overlapping tasks across packages render on distinct lanes. The shared pid: 1 is arbitrary.

cat carries the task’s final status (e.g. cache-hit, failed) so trace viewers can color-code.

Errors

Both writers throw on FS failure. The orchestrator catches and surfaces via log.status but does NOT fail the run — the work already happened; an inability to write an audit artifact shouldn’t change the exit code.

Tests

tests/run-artifacts.test.ts covers:

• Summary written to default path when --summarize has no value.
• Summary written to explicit path when --summarize=foo.json.
• Profile written and parseable as JSON.
• Hrtime bigints preserved as strings.
• Profile writers skip tasks without hrtime spans.
• Group tasks excluded from both outputs.

Replacing this module

The two writers are independent. To replace --profile output (e.g. emit OpenTelemetry instead), swap writeRunProfile and keep SummarizeArgs / ProfileArgs shapes. The orchestrator only knows about the two target fields.