Skip to content
Back to skills

Dashboards

ASecurity

Diagnose empty dashboards by replaying console panels and separating absent data from window, scope, filter and query failures.

  • 10 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added November 13, 2026
ai-agentsbashsqldebugging

Security analysis

A100/100

Scanned November 13, 2026

npx -y skills add reliant-labs/forge --skill dashboards --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Dashboards?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Dashboards
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/reliant-labs-dashboards-forge/badge)](https://www.skillsdirectory.com/skills/reliant-labs-dashboards-forge)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: dashboards
description: Diagnose empty dashboards by replaying console panels and separating absent data from window, scope, filter and query failures.
---

# Debugging empty dashboards

#### An empty dashboard: do we have the data, and why isn't it showing?

Two different questions, and one command answers both. `forge obs dashboards
audit` lists the dashboards the console lists for the env and sends each panel
to the same server-side `QueryPanel` the console calls, over the range the
console opens each dashboard on (its `defaultRange` rounded up to a preset, else
the last hour; `--since` replaces it, as a viewer who picked a range), with the
step left to the control plane. A sql panel whose `minStep` is wider than that
window is widened the way the console widens it (to the smallest preset holding
two buckets), and the panel reports `widened` and the window it really ran over.
It rebuilds no panel SQL, so it cannot disagree with the console.

```bash
forge obs dashboards audit prod                                  # every panel, failures first
forge obs dashboards audit prod --dashboard product-usage --json  # one dashboard, as a document
forge obs dashboards audit prod --dashboard product-usage --panel workflows-by-outcome --since 24h
```

Read each panel's `status`, `reason` and `probable_cause`, not just the status:

| status | meaning | next |
|---|---|---|
| `ok` | data the console can draw (`all_zero` in `note` means it is zero, not missing) | none |
| `partial` | data, but a series' metric is absent or stored as another kind, or the answer is **one bucket** where a line needs two *under the console's own window* (a day-bucketed sql panel with no `minStep`, which the console cannot widen) | declare `minStep` / `defaultRange`, or fix the panel's bucketing |
| `empty` | nothing came back; `probe.verdict` says whether the data exists | by verdict, below |
| `error` | the read failed; `error.class` names it | by class, below |

An empty panel is probed over `--probe-window` (default 7d): the same panel in
every env of the project, and without its `where` filters, plus a metric catalog
lookup. The verdict separates **no data** from **data outside the window, scope
or filter**. A panel whose window is already the longest its signal allows (a 7d
traces dashboard) cannot look further back; the env, filter and catalog probes
still run over its own window, and the skipped step says why:

- `data_outside_window`: the data exists; the window is shorter than its cadence
  or the emitter went quiet. Widen `--since`.
- `data_in_other_env`: it exists in another env; rows carry that env's id, or none
  (an env-scoped panel cannot see untagged rows).
- `data_excluded_by_filters`: it exists without the panel's `where`; the filter
  value matches nothing the emitter writes.
- `data_in_other_env_without_filters`: it exists only with both lifted: another
  env, and the filter matches nothing there either.
- `metric_never_seen` / `no_data_anywhere`: the emitter is not shipping it, or the
  name is wrong. `forge obs metrics describe <env> <metric> --since 168h`, then
  `forge obs query` for the table itself.
- `probe_failed` / `not_probed`: the follow-up could not look (a read failed, or
  nothing was left to check); this is **not** "no data". Fix the cause in
  `probe.steps[].error`.

Failure classes are never "no data": `auth` (HTTP 403 or a missing scope such as
`obs:read`: a credential problem for an org admin, not an outage), `quota` (the
org's hourly allowance is spent; the audit stops reading and says so), `in_flight`
and `rate_limited` (retried, then reported; wait and rerun, or audit one
`--dashboard` at a time), `timeout`, `storage`, `invalid` (a spec or range the
control plane refused). Exit 0 means every panel is `ok`, **2** that some panel is
empty or partial (or the audit could not look), **1** that a panel or dashboard
could not be read. A local env has no dashboards to replay: use
`forge obs dashboards check --against <env>` for its metric names.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

View all in ai-agents →

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.