Skip to content
Back to skills

Common Telemetry

ASecurity

Enforce session-cost telemetry and execution-metadata reporting. Use when explicitly invoking get_session_cost, reporting token/cost usage, applying telemetry or cost guidance during a workflow handoff, or writing artifacts/session-cost.md at a workflow terminal state.

  • 549 stars
  • 0 votes
  • 0 copies
  • 2 views
  • Added June 6, 2026
development

Works with

  • terminal
  • mcp

Security analysis

A100/100

Scanned September 24, 2026

npx -y skills add HoangNguyen0403/agent-skills-standard --skill common-telemetry --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Common Telemetry?

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

Security grade badge for Common Telemetry
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/hoangnguyen0403-common-telemetry/badge)](https://www.skillsdirectory.com/skills/hoangnguyen0403-common-telemetry)

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: common-telemetry
description: Enforce session-cost telemetry and execution-metadata reporting. Use when explicitly invoking get_session_cost, reporting token/cost usage, applying telemetry or cost guidance during a workflow handoff, or writing artifacts/session-cost.md at a workflow terminal state.
metadata:
  triggers:
    files: []
    keywords:
      - token cost
      - token usage
      - session telemetry
      - cost report
---

# Telemetry & Cost Reporting

## **Priority: P2 (MEDIUM)**

## 1. Finalizing a Workflow

As your final step in any SDLC workflow (or when a user explicitly requests session cost):

1. Call the `get_session_cost` tool provided by the agent-skills-standard MCP server.
2. Pass `workflow`, `model`, token counts, cache/reasoning usage, and per-1M token rates when the host runtime exposes them. Set `slug` (feature/workspace name) and `outcome` (terminal `feature_status`) when this call finalizes a workflow, so the telemetry log can attribute cost per feature.
3. Never set `costSource: "host"` yourself — that value is reserved for the runtime's own telemetry adapter (`buildSessionCostRequest`). If you type numbers by hand, omit `costSource` so the tool reports them as `agent-estimate`, never `host`.
4. If token counts are unavailable, report MCP-observed telemetry and let `get_session_cost` mark cost provenance as `unavailable` — never fabricate a number or a `$0.00` total.
5. Append a Markdown table containing the usage metrics to `artifacts/session-cost.md`.

When asked which artifact to use, always name both `artifacts/session-cost.md` and `get_session_cost`; if the host does not expose the helper, say so explicitly while retaining the artifact requirement.

## 2. Host Runtime Contract

- This skill defines the telemetry contract; it does not collect provider billing data by itself.
- The host runtime or orchestrator must trigger the final telemetry call when the workflow reaches a terminal state such as `completed`, `failed`, or `blocked`.
- The host may use a reusable helper such as `mcp/src/services/WorkflowTelemetry.ts` to:
  - decide when telemetry should fire
  - build the `get_session_cost` payload
  - pass prompt, cache, reasoning, pricing, and other runtime cost fields when available

## 3. Telemetry Format

Ensure the `artifacts/session-cost.md` or the output template `## Cost Report` follows this structure, copying the `get_session_cost` output verbatim rather than re-deriving values:

| Metric                   | Value                                                  |
| ------------------------ | ------------------------------------------------------- |
| **Tool Calls**           | [from get_session_cost]                                |
| **Skills Loaded**        | [from get_session_cost]                                |
| **Workflows Loaded**     | [from get_session_cost]                                |
| **Prompt Tokens**        | [from get_session_cost, or "unavailable" if not exposed] |
| **Cached Prompt Tokens** | [from get_session_cost, or "unavailable" if not exposed] |
| **Completion Tokens**    | [from get_session_cost, or "unavailable" if not exposed] |
| **Reasoning Tokens**     | [from get_session_cost, or "unavailable" if not exposed] |
| **Other Runtime Cost**   | [from get_session_cost, or "unavailable" if not exposed] |
| **Cost Provenance**      | `host` \| `agent-estimate` \| `unavailable`              |
| **Estimated Cost**       | [from get_session_cost, or "unavailable" — never `$0.00`] |

## Anti-Patterns

- **No skipping the telemetry step**: Always include the Cost Report at the end of the execution if mandated by the workflow.
- **No fabricated numbers**: Never invent a `$0.00` or any other placeholder cost when `get_session_cost` reports `unavailable`; render the provenance value as-is.

## Canonical response anchors

When this skill applies, preserve the following domain terminology or equivalent concrete examples in the answer when relevant:
- Call `get_session_cost` when the host exposes it, then append the Markdown table to `artifacts/session-cost.md`.
- Cost Provenance: `host`, `agent-estimate`, `unavailable`

- Telemetry & Cost Reporting

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.

Comments

Loading comments…

Newsletter

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