The sprint, seen whole.

Beyond the features you described, these twelve requirements are what would make ScrumBoost feel unique rather than "another Jira dashboard." Each is annotated in the screens below.

The full vision is several products in one — auth, connectors, analytics, an agent runtime, voice. Building it in slices keeps each implementation plan small enough to execute well (each phase gets its own subagent-driven plan). The spec locks M1–M4 as internal build order only: nothing is cut, and all four ship together as v1.0.

Quiet and fast — the product's personality shows on the brand side, the form stays frictionless. One account works on both web and desktop.

1. 1Email + password, or GitHub/Google SSO. GitHub SSO doubles as a head start for the GitHub connector on the next screen. Registration asks only email, password, and workspace name — everything else waits for onboarding.

The onboarding wizard is where trust is won: connect Jira, Confluence, and GitHub, then declare exactly whose data may be pulled. The scope preview makes "we don't take more than you gave" visible before the first sync runs.

1. 1Per-tool connection cards — Jira and Confluence connect via Atlassian OAuth with read-only scopes; GitHub is a GitHub App installation on explicitly selected repositories (read-only permissions, 1-hour tokens — no PAT field ships). Each card shows exactly what is in scope; the exact scope list is printed before consent, and scope is editable any time in Settings. Slack is not here: it's an outbound-only publish target configured later in Settings, never an ingestion source.
2. 2Roster-scoped ingestion — you map each teammate's Jira username and GitHub handle once; sync queries filter by these identities so no out-of-scope data is pulled.
3. 3Scope preview before first sync — the system states what it will fetch and how long it will take, which becomes the permanent sync-status bar after onboarding.

The daily home screen: a widget grid you can rearrange to match how you run the day. Ceremony presets swap the whole layout in one click.

▶ This mock is live — grab a widget by its header and drag it onto another to rearrange (mouse/trackpad).

1. 1Draggable widget grid — rearrange, resize, add/remove widgets; layouts persist per user, and ceremony presets (Planning / Standup / Retro) swap the arrangement in one click.
2. 2Sprint Health Score — the composite 0–100 metric with its contributing factors as chips; clicking a chip drills into the underlying data.
3. 3Difficulty heatmap — every issue colored by measured effort (damped churn, files, commits, review rounds & comments, cycle time), so hard spots are visible before they burn the sprint. Effort describes issues, never people.
4. 4Blocker radar — items stuck beyond team norms, with the reason when detectable; this list feeds the Insights suggestions.
5. 5Ingestion transparency — always-visible sync freshness per connector; clicking opens sync logs and scope settings.

Everything the team did on GitHub — commits, pushes, PRs, reviews, comments — rolled up per member and streaming as a live feed, always linked back to the Jira story.

1. 1Per-member rollups — commits, PRs opened/merged, reviews given, and comments for the selected sprint and repos. Clicking a row opens that member's story threads, not a surveillance profile.
2. 2Privacy guardrail in action — the default is team-aggregate, and any aggregate covering fewer than three people is suppressed (k ≥ 3) instead of narrowing to one person. Enabling individual view takes Owner/Admin step-up re-auth plus a stated purpose, auto-expires after 8 hours, and the badge renders from a server response envelope — data can't flow without it. Enable, expiry, and every read are audit-logged; there are deliberately no leaderboards.
3. 3Live feed with story links — every GitHub event is matched to its Jira issue (branch name, PR title, commit message); clicking opens the full story thread timeline.
4. 4Low-confidence links ask first — story links at confidence ≥ 0.70 render as active; 0.50–0.69 appear only as possible matches with Confirm / Reject. A weak guess is never silently asserted, and your decision is sticky — recompute never overwrites it.
5. 5External people are pseudonymised — a reviewer outside the roster (here on PR #482) still counts toward review latency, but their name, handle, and avatar are never stored: they render as External reviewer · ext:9f3a1c, a workspace-salted hash. The roster bounds people, not just repos.

The calibration engine compares story points with measured effort — a damped composite of code churn, files, commits, review rounds & comments, and cycle time, computed after stripping generated, vendored and lockfile diffs — then turns what it finds into suggestions that always show their evidence. Effort is a signal about where work was hard, never a rating of a person.

1. 1Calibration scatter — each dot is an issue; the green band is your team's fitted points-to-effort curve, refit from sprint data as work closes (never from feedback). Effort excludes generated, vendored and lockfile diffs, cancels moved code, log-damps churn so one refactor can't dominate, and counts deletions as work — a signal for the Scrum Master, not a score on a person.
2. 2Evidence-first suggestions — no black-box advice: every card lists the exact signals behind it. Accept/Adjust/Dismiss tunes exactly one named threshold per suggestion type — deterministic, bounded to ±40%, stored as inspectable per-team settings, resettable in one click. No model is trained; nothing crosses tenants.
3. 3Hand-off to the agent — any suggestion can be handed to the agent to execute as a workflow (draft the issue split, propose reviewer reassignment) — which then goes through approval mode.

A Claude-powered copilot that lives beside the data. You talk (or type); it answers with evidence, builds workflows you can see, and never writes anything back without approval. Anthropic is the launch provider — the provider layer is pluggable from day one.

1. 1Provider layer — Anthropic first via the Claude Agent SDK, authenticated with an Anthropic API key. Subscription-token mode exists but ships behind a flag, off by default — its terms approval is an open launch decision (footer). The interface is provider-agnostic so others can be added post-v1.0 without UI changes.
2. 2Approval mode — any side effect (creating a workflow, posting to Slack, commenting on Jira) is proposed as a card with a full preview. Approve, edit, or reject. Slack itself is outbound-only: a publish target configured in Settings with allowlisted channels, never an ingestion source.
3. 3Live workflow canvas — the agent builds the workflow visually as it reasons; you can edit steps directly. Saved workflows appear under the Workflows tab and can run on schedules or triggers.
4. 4Chat + voice — same conversation, two inputs. Voice is push-to-talk first (reliable, private); wake-word can come later.