You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
[RSI Baseline] Build a project dashboard for autonomous engineering #790
The existing Bot Timeline shows team execution and utilization, while the Agent Inbox proposal focuses on tasks, approvals and failures. RSI needs a project-level view of delivery outcomes and process bottlenecks, backed by durable task and GitHub evidence.
Scope
Unified project board for Issues/tasks, active runs, worktrees, PRs, review, CI, merge queue, releases and blocked/failed work.
Show current state, Bot/owner, elapsed time, last activity, retry count, blocking reason and evidence links.
Show task-level and aggregate metrics: cycle time, queue/wait time, first-pass review rate, CI pass rate, rework, defect escape rate, human intervention, and estimated model/Runner cost.
Distinguish observed facts from missing or estimated data.
Filter by time range, repository, capability-pack version and Code Agent/model.
Provide a machine-readable API/export for RSI evaluation; avoid exposing secrets or raw task contents by default.
Acceptance criteria
Dashboard values are traceable to durable event records and GitHub/Runner sources.
Stale or missing telemetry is marked as such, not silently interpreted as zero.
A user can inspect one task end-to-end and identify where it spent time and why it is blocked.
Existing Timeline and Agent Inbox features are reused where appropriate rather than duplicated.
enhancement|area:ui|area:automation|needs-human. Three existing things cover parts of this and two were built to do exactly the first bullet, so this has to say what it adds:
#349 (closed) was the Agent Inbox board — "one place for tasks, approvals", and #471 (open) is its phase 2: board view, tool-call counts, failed turns and team views. A new project board that lists tasks and approvals without saying how it differs from #471 is a duplicate.
#582 (closed) is the Timeline team run report and #584 (closed) its durable alerts; #357 (open) is the Nightly CI metrics dashboard.
The data layer exists: .github/agent/ state, the controller's records, and the GitHub API as source of truth.
What is genuinely missing is the project-level view (cycle time, queue wait, first-pass review rate, rework, human intervention, cost) with "observed vs missing" distinguished — that is a metric definition problem before it is a UI problem, and it depends on #791. Building the surface first would produce numbers with no defined source.
To become agent-ready: one screen's worth of fields, each with a named source (which endpoint, which file), and the rule that missing telemetry renders as "no data" rather than zero — a test that goes red when an absent metric is drawn as 0.
Closing on the owner's call this round, folded into #471. The board view this asks for is #471's phase 2 (Agent inbox phase 2: board view, tool-call counts, failed turns and team views); #349, which it would extend, is already closed and #471 keeps the history and the acceptance. Two open issues for one board produce two boards. File again only if the dashboard has to be something other than the inbox board.
Activity
Birfy commentedon Oct 22, 2026
Written-by: Triage
enhancement|area:ui|area:automation|needs-human. Three existing things cover parts of this and two were built to do exactly the first bullet, so this has to say what it adds:#349(closed) was the Agent Inbox board — "one place for tasks, approvals", and#471(open) is its phase 2: board view, tool-call counts, failed turns and team views. A new project board that lists tasks and approvals without saying how it differs from#471is a duplicate.#582(closed) is the Timeline team run report and#584(closed) its durable alerts;#357(open) is the Nightly CI metrics dashboard..github/agent/state, the controller's records, and the GitHub API as source of truth.What is genuinely missing is the project-level view (cycle time, queue wait, first-pass review rate, rework, human intervention, cost) with "observed vs missing" distinguished — that is a metric definition problem before it is a UI problem, and it depends on
#791. Building the surface first would produce numbers with no defined source.To become
agent-ready: one screen's worth of fields, each with a named source (which endpoint, which file), and the rule that missing telemetry renders as "no data" rather than zero — a test that goes red when an absent metric is drawn as0.Birfy commentedon Oct 23, 2026
Closing on the owner's call this round, folded into #471. The board view this asks for is #471's phase 2 (
Agent inbox phase 2: board view, tool-call counts, failed turns and team views); #349, which it would extend, is already closed and #471 keeps the history and the acceptance. Two open issues for one board produce two boards. File again only if the dashboard has to be something other than the inbox board.