CoIsland use cases for your team

CoIsland is a notch app for work. Each monitor is a ring counting down to its next check, with its match count under it. Click a ring to open its newest alert; press and hold it to mark what is new as seen. The home page shows the notch cards.

Alert triage with AI agents happens right there: the agent runs its own CLI in a terminal under the alert, and you answer its plan. Agents covers all three.

Role	Monitors to start with
Data engineer	Custom SQL (Snowflake) for failed tasks; Failed workflow runs (GitHub) on the pipeline repository.
Analytics engineer	Validation (Snowflake) for nulls, out-of-range values or a status outside a set; Review requests (GitHub).
Software engineer	Review requests and Failed workflow runs (GitHub); Issues (Jira) for assignee = currentUser() AND statusCategory != Done.
Tech lead	Security alerts and New releases (GitHub); Sprint scope (Jira).
On-call	Assigned to me and Shift starting soon (PagerDuty); Failed workflow runs (GitHub) on the deploy workflow.

Each connector page lists its monitors, and Connectors has them all side by side. The Monitors page explains the rules they share.

CoIsland has no team account and no server: each person runs it on their own Mac, with their own connectors. What a team shares is its monitors, because each one is a plain text file.

Monitors live in ~/.coisland/watches, one .sql file each: settings in -- key: value comment lines, then the query. GitHub and Jira monitors add a -- kind: line before their search or JQL.

-- name: Failed tasks
-- connector: analytics
-- database: ANALYTICS
-- every: 5m
-- alert: new-rows
-- key: QUERY_ID
-- title: {NAME} failed: {ERROR_MESSAGE}
SELECT query_id, name, error_message, scheduled_time
FROM TABLE(INFORMATION_SCHEMA.TASK_HISTORY(
    SCHEDULED_TIME_RANGE_START => DATEADD('hour', -1, CURRENT_TIMESTAMP())))
WHERE state = 'FAILED'

• Keep the folder in git. Only the visible .sql files at its top count, so .git and a README are ignored. A file you pull shows up within a second.
• No secret in the file. It names a connector, or none for your default one; each teammate adds that connector with their own token.
• New-rows monitors start quiet. The first check only records what already matches. A count monitor already past its line alerts at once.
• Alerts stay on each Mac, with their notes, agent sessions and dashboards. There is no shared inbox.

Watch files lists every header key.

A monitor that alerts all day teaches people to ignore the notch. Dashboards show, over 24 hours, 7 or 30 days, alerts raised and open, the share someone handled (a falling share is alert fatigue), and the median time to acknowledge and to resolve. Filter by connector, then tighten or retire the noisiest monitors.

A page for support teams is next. Today, the Jira connector's SLAs monitor already watches Jira Service Management requests that breach an SLA, or are about to:

-- name: Breached requests
-- kind: jira.sla
-- every: 5m
project = HELP AND "Time to resolution" = breached()

Each request alerts when it starts matching.