Dashboards

Colony Cloud provides operator surfaces under the Dashboard group in the left navigation. This page documents Pipeline, Insights, and Inbox in full. For the ambient Live view (/dashboard/pipeline-live) and the Reviews inbox, see their dedicated pages: Live and Reviews.

Pipeline

Route: /dashboard/pipeline

Pipeline is the first screen after sign-in and the primary daily-use surface. It shows active work flowing through stations, organized per repository, as a real-time swim-lane board.

Each repository gets its own swim lane. Within the lane, issues appear as cards at their current station — analyzing, planning, in-review, and so on. The view updates on a live connection; no manual refresh is needed.

Tip

Sort by age descending to surface issues that have been in the same state for an unusually long time. Issues that have not transitioned in more than a few hours are candidates for operator intervention.

For a state-machine grid showing every pipeline state as a count node — and list and board views scoped to individual states — see the Pipeline Live surface at /dashboard/pipeline-live.

Insights

Route: /dashboard/insights

Insights shows engineering leadership KPIs at the top and cycle-time forensics below. It is the right surface for weekly reviews, sprint retrospectives, and cost tracking conversations with stakeholders.

Leadership

The Leadership section answers three questions for the selected 7-day or 30-day window:

Question	Metric	What it shows
01 — Are we shipping?	Throughput	Issues completed (PR merged) within the selected window, with a daily bar chart comparing the current half of the window to the prior half.
02 — Is quality holding?	PR Revert Rate	Fraction of merged PRs subsequently reverted. Spikes here warrant a review of issue quality and review strictness.
03 — Are we efficient?	Autonomous Merge Rate	Fraction of merged issues completed end-to-end by Colony with no human commits and no changes-requested cycle. When autonomy data is unavailable, the card falls back to Agent Attribution — the share of merged PRs authored by Colony versus humans.

Each question card shows a delta badge comparing the current window to the prior equivalent window. Green indicates improvement; amber or copper indicates a meaningful regression worth investigating.

Cost & Efficiency

Model-level spend breakdown:

Metric	What it shows
Model Spend Share	Fraction of total org LLM spend attributable to each model in the selected window. A disproportionately large share from one model may indicate over-use.
Max-Turns Rate	Fraction of invocations for each model that hit the agent turn limit. A consistently high rate suggests tasks may be too large for that model’s context window.
No-Progress Rate	Fraction of invocations for each model that produced no meaningful code progress. A high rate suggests the task type or prompt structure is not well suited to that model.

For the full cost story — multi-horizon spend totals, spend timeseries, Budget Overview (burn rate and projected exhaustion), and Monthly Cost Projection — see Cost & Budget.

Forensics

Cycle-time forensics for issues that entered the pipeline in the selected window:

What it shows	Details
Cohort funnel	For each spine state (New → Analyzing → Planning → Ready for dev → In review → Merge pending → Done), shows the count of issues that reached that state and their median dwell time. Long dwell times identify bottlenecks — drill into Pipeline Live to investigate.
Off-spine drops	State transitions that leave the main pipeline path (for example, into failure-blocked, dependency-blocked, or paused) listed as badge pairs with the count of issues affected.
State Timeseries	Time-series chart of issue counts by pipeline state over a selectable window (1h / 6h / 24h / 7d).

Pipeline Health

Regression signals comparing the rolling window against a baseline:

Metric	What it shows
Block Rate	Fraction of issues that transitioned to a blocked state, relative to all issues active in the window. A rising rate signals increasing pipeline impedance.
Failure-Blocked Rate	Fraction of issues that entered failure-blocked, relative to all issues active in the window. A rising rate usually indicates flaky tests or environment instability.
Mean Cost / Issue	Mean LLM token spend per completed issue. A rising mean indicates the pipeline requires more LLM work per deliverable.
Mean Turns / Issue	Mean number of agent turns consumed per completed issue. A rising mean signals increased difficulty or iterative rework.
Reimplement Loop Rate	Fraction of issues that triggered a re-implementation cycle. A rising rate points to planning prompt quality or codebase comprehension gaps.
Dead Letter Rate	Fraction of terminal issues associated with dead-lettered transitions. A rising rate requires manual triage; use issue context plus the authenticated dead-letter API procedure.

Inbox

Route: /dashboard/inbox

Note

Both /attention (without the /dashboard/ prefix) and /dashboard/attention are permanent redirects to /dashboard/inbox. Bookmarks and external links using either short form continue to work.

Inbox surfaces issues that require operator action — the subset of the pipeline that the autonomous recovery protocols could not resolve on their own. Issues in the Needs you section are sorted by urgency: blocked and awaiting-external first, then awaiting operator action, then review-ready, then issues that are merely stuck.

Inbox sections

Section	What appears here
System Alerts	LLM health warnings and operational alerts affecting the fleet as a whole.
Strategy proposals awaiting review	Strategist-proposed snapshots that need operator approval before becoming active.
Recurring failures	Error patterns appearing across two or more issues — a signal of systematic problems in conventions, CI, or environment configuration.
Needs you	Individual issues that have stalled, are blocked, are awaiting human code review, or need other direct operator intervention.

Common interventions

From the Inbox, you can act directly on each issue:

Situation	Action
Worker stalled or timed out	Click Retry to clear the block, reset failure counters, and re-enqueue. A sub-menu lets you optionally select a complexity-tier override (equivalent to /colony:retry [size]).
Issue is blocked on a dependency that has since resolved	Click Unblock to remove the current block and re-enqueue.
Review loop caused by a bad issue or bad conventions	Fix .colony/conventions.md, then click Retry.
Issue should be abandoned	Click Cancel to transition to Done and close the GitHub issue.
Issue has related dead-letter context	Diagnose the issue, then use Recover a Dead-Lettered Issue when the record itself must be retried or resolved.
Processing should halt temporarily without losing state	Click Pause to pause processing while preserving current state.
Issue was previously paused and should resume	Click Resume to unpause and restore to the pre-pause state.
Analysis is complete and planning should be skipped	Click Skip planning to go straight to development.
Issue scope is too large and should be broken into subtasks	Click Decompose to force epic decomposition into subtasks.
Current implementation is wrong and should be restarted from scratch	Click Reimplement to close the PR, delete the branch and worktree, and reset to analyzing.
PR is ready for a re-review cycle	Click Trigger review to start a new review cycle.
PR needs targeted changes the automated reviewer cannot capture	Click Return with feedback to send change-request text back to the developer agent.
Pipeline state is incorrect and must be corrected manually	Click Override state to manually set the pipeline state to any valid state.
Issue priority needs to change	Click Set priority to set priority to high, normal, or low.

For a full list of slash commands available on individual issues and PRs, see the Operator role.