Skip to content

monitoring-and-artifacts.md

Latest commit

 

History

History
117 lines (91 loc) · 4.95 KB

monitoring-and-artifacts.md

File metadata and controls

117 lines (91 loc) · 4.95 KB
Raw

Monitoring And Artifacts

Quay read commands emit deterministic JSON on stdout. Write-command failures emit JSON on stderr.

List Tasks

quay task list
quay task list --state queued
quay task list --state queued --state running
quay task list --repo myrepo
quay task list --external-ref ENG-1234

Output is a JSON array ordered by creation time and task id. Rows include:

  • task_id
  • repo_id
  • base_branch
  • state
  • external_ref
  • branch_name
  • attempts_consumed
  • retry_budget
  • budget_exhausted
  • created_at
  • updated_at

Get One Task

quay task get <task_id>

Output includes the task-list fields plus:

  • pr_number
  • pr_url
  • head_sha
  • base_sha
  • current_attempt
  • recent_events

recent_events is limited to the latest 20 events.

Full Event Log

quay task events <task_id>

Output is the full append-only task event log, oldest first.

Artifacts

Get the latest artifact of a kind:

quay artifact get <task_id> brief
quay artifact get <task_id> blocker
quay artifact get <task_id> session_log

Get an artifact for a specific attempt:

quay artifact get <task_id> brief --attempt 2

Print only the artifact path:

quay artifact get <task_id> final_prompt --path

Without --path, Quay streams raw artifact bytes to stdout. This preserves binary and invalid UTF-8 artifacts such as malformed_signal.

Common Artifact Kinds

Kind Meaning
ticket_snapshot Snapshot of source ticket/context. Enqueue writes the first task-level row; task resnapshot can append a newer task-level snapshot after the live ticket changes.
task_objective Canonical task brief reused by later code-worker and reviewer prompts. Enqueue writes the first task-level row; task resnapshot can append a newer task-level objective.
brief Per-attempt composed prompt body: a structured <quay-task-objective> block pointing at the current task-level task_objective artifact, a <quay-current-attempt-guidance> block (initial instruction, retry/respawn template, or orchestrator-submitted brief), and an optional <quay-diagnostics> block (CI excerpt, review comments, conflict slice, etc.). The current task objective lives in task_objective, not here.
final_prompt Code-worker attempts: worker preamble plus the attempt's composed brief. Review attempts: static reviewer protocol, reviewer guidance, then the review brief.
session_log Captured tmux output.
usage JSON usage envelope captured per attempt. .quay-usage.json is stored verbatim when present; otherwise Codex --json JSONL in .quay-tool-trace.log can synthesize normalized model/token totals.
tool_trace Raw tool-call/debug stream or Codex JSONL from .quay-tool-trace.log, captured per attempt with a 4 MiB tail.
blocker Valid .quay-blocked.md written by a worker.
adopted_pr_ready_for_review Valid .quay-ready-for-review.json written by an adopted PR worker to report that the PR already satisfies the requested work and should return to normal PR polling/review without a new commit.
malformed_signal Invalid blocker signal bytes.
goal_report Valid .quay-goal-report.json captured from a goal-mode worker.
malformed_goal_report Invalid goal report bytes captured for bounded protocol repair.
goal_evidence File, URL, artifact-reference, or note evidence cited by a goal report.
goal_evidence_manifest Manifest linking each cited evidence entry to its captured artifact or validation error.
goal_completion_audit Durable audit decision for a complete goal report, including reasons and follow-up feedback.
ci_failure_excerpt CI failure details captured from GitHub checks.
review_comments Snapshot of posted review body and inline comments. Used for review respawns and synthetic review forensics.
review_blocker Reviewer blocker, review-context-drift diagnostic, or infrastructure-failure diagnostic.
conflict_slice Snapshot of merge-conflict observation.
slack_escalation_post Human question body recorded by escalate-human.
slack_reply Human Slack reply recorded by the orchestrator, or by legacy tick ingestion.
last_failure Would-be retry brief when retry budget is exhausted, or a terminal worker-auth preflight diagnostic after the one fresh-auth retry also fails.

Tick Errors

Tick isolates per-task failures. When one task fails during a tick, Quay records a tick_error event and continues processing other tasks. A later successful tick path clears the task's tick_error field.

Worker GitHub auth preflight failures are not tick_error: they are classified as worker_auth_invalid, retried once after spawn_retry_next_eligible_at with freshly resolved credentials, then surfaced through an awaiting-next-brief handoff if the retry also fails. Spawn/reviewer infrastructure failures record their latest operator-facing reason in tasks.spawn_failure_reason.