Token Observability
The token-observability skill helps you understand and manage AI token usage
in your project — surfacing costs, tracking consumption, and setting budgets.
Install the skill
ag-skill install token-observability
What the skill provides
- Guidance for logging and monitoring token usage per agent call
- Patterns for tracking cumulative costs across a session
- Recommendations for setting context-window budgets
- Tips for reducing unnecessary token consumption
Usage
Once installed, your AI agent will have access to the token observability guidance automatically in context. Reference the skill when asking your agent to help optimize or audit token usage in your workflows.
For full details, see the Token Observability Skill Guide in the ag-skill repository.
Built-in telemetry commands (v1.0.1+)
ag-skill v1.0.1 introduced CLI commands for recording and reviewing skill/agent run
telemetry locally — no external service required. This complements the
token-observability skill by giving you a persistent, queryable run log.
| Command | Description |
|---|---|
ag-skill observe | Display recent telemetry records |
ag-skill observe record | Append a run record (tokens auto-estimated or supplied exactly) |
ag-skill observe from-copilot | Auto-read the latest VS Code Copilot Chat transcript |
ag-skill observe from-claude | Auto-read the latest Claude Code session (exact token counts) |
ag-skill stats | Aggregated statistics — totals, per-skill, per-model |
ag-skill dashboard | Generate a self-contained HTML dashboard |
See the Local Telemetry guide for the full workflow and all options.
What changed in v1.1.0
v1.1.0 expanded the token-observability and related standards skills with richer
references and script assets. It also standardized generated report paths to use
the .ag-skill folder convention.
Preferred report output location:
.ag-skill/reports/token-observability-<timestamp>.md