# Skills

Inventory the skills your agent loads, then compare their usage, completion, cost, hand-offs, reloads, and tool failures.

Brizz's **Skills** surfaces show which named skills your agent loads and whether they help sessions finish successfully. Use the inventory to inspect one skill; use the dashboard to compare skills across a service.

## How skills appear

Skills are extracted from your service's telemetry. If the service has no detected skills, [contact us](mailto:support@brizz.ai) so Brizz can be configured to recognize them.

A detected skill can include its name, description, full manifest, declared tools, and manifest fingerprint. Fields that were not captured stay empty rather than being inferred.

## Inspect one skill

Open **Observe → Skills** for the service's all-time skill inventory. You can search by name, sort the supported columns, and open a row to read its manifest and observed usage.

The table shows:

- **Completed outcome** — completed sessions divided by sessions with a recorded outcome. A dash means no outcome data, not 0% completion.
- **Loads and sessions** — loads count every time the skill was loaded; sessions count distinct sessions. Loads can therefore exceed sessions.
- **Reload** — the share of sessions with skill-load telemetry that loaded the skill at least twice.
- **Active issues** — distinct open or tracked issues in sessions that used the skill. This is association, not proof the skill caused them.
- **Tools** — declared tools reconciled with observed calls, including a high-volume failing tool or a declared tool that was never observed.
- **Last seen, manifest size, and last changed** — manifest history is available only when fingerprinted skill-load telemetry was captured.

Click the session or issue count to open the corresponding filtered list. Click anywhere else on the row to inspect the description, manifest, declared and observed tools, outcome, reload rate, and first/last-seen timestamps.

## Compare skills

Open **Dashboards → Skills** and choose a date range. Counts cover sessions that started in that range, while the full vocabulary remains visible so an inactive skill can still appear at zero.

### Usage

Use the load trend and ranking to see whether traffic is concentrated in a few skills. **Dead & barely-used skills** surfaces skills with at most 10 sessions or no load in the last 30 days.

**Skills loaded back-to-back** counts consecutive skill loads inside a session. One-way chains, traffic that returns, loops, and fan-out are shown separately. If two loads share a timestamp, Brizz uses a deterministic tie-break; the row calls this out when tied events are a material share of the pair.

### Quality

Completion is calculated only from sessions with a recorded outcome. Comparisons require at least 30 scored sessions per skill; the typical value is the median across qualifying skills.

The reload finding compares sessions that loaded a skill once with sessions that loaded it two or more times. It appears only when both groups have at least 30 scored sessions, the completion difference is statistically significant, and reloaded sessions complete less often. An absent card is not a count of reloads; use the inventory's **Reload** column for the plain rate.

### Cost

The cost section compares total spend, cost per session, and cost per success. **Cost per success** divides spend across all scored sessions by the number that completed, so failed attempts remain part of the price.

A session's whole bill is credited to every skill it loaded. Co-occurring skills can therefore show identical costs, and per-skill totals can add up to more than the service total. Treat these figures as attribution for comparison, not as an allocation that reconciles to service spend.

### Tools

The tools section separates a broadly failing shared tool from a failure localized to one skill. Tool calls are attributed to the most recently loaded skill in the session. A failure rate is shown only after at least 20 attributed calls, and the dashboard flags rates of 5% or higher.

## Availability

The Skills dashboard is off by default and must be enabled for your organization. If either Skills entry is missing from navigation, [contact us](mailto:support@brizz.ai).

## See also

- [Sessions](/docs/platform/observe/sessions.md) — filter sessions by skill and inspect the underlying conversation.
- [Issues](/docs/platform/issues.md) — triage the active issues associated with a skill's sessions.
- [Tools](/docs/platform/dashboards/tools.md) — investigate one tool across the service.
- [Cost](/docs/platform/dashboards/cost.md) — reconcile service-level model spend.