DevOps Agent Cost Dashboard¶

View on GitHub

This custom agent runs the devops-agent-cost-insights skill on a schedule, scoped to one IAM role, and emits a fixed dashboard plus a text summary. Where the skill is the method, this agent is the persona that runs it: it pins the configuration, fixes the reporting window, executes silently, and produces the same report layout on each run.

Use it when you want a recurring, charted cost report for a single DevOps Agent role rather than an ad-hoc chat answer.

Figures are estimates, not billing data. They are reconstructed from CloudWatch metrics, CloudTrail events, and query execution history, priced at published list rates. AWS Billing and AWS Cost Explorer remain the authoritative record of what you were charged. This output must not be used as the basis for invoicing, chargeback, or contractual commitments.

Architecture:¶

The scheduled cost dashboard agent applies the devops-agent-cost-insights skill, which reads CloudWatch metrics, the Pricing API, Logs Insights and Athena query history, CloudTrail, and journal records, then emits a charted dashboard artifact and a fixed text summary

What it produces¶

Each run creates or refreshes a single artifact named "DevOps Agent Cost Dashboard" and then emits one text summary. Both cover the same trailing window:

Direct usage, per day, priced per agent-second and scoped to the Agent Space
Credits, shown as an account-level reference (see the scope note below)
Downstream AWS API cost, attributed to the one configured role
Tool usage, including Azure and third-party MCP tools that do not appear on an AWS bill
Validation notes, including a zero-count attestation for each paid operation the run is forbidden to issue

The report keeps direct usage (space-scoped) and downstream (role-scoped) as separate figures.

Generated output¶

Each run emits two things: a charted dashboard artifact and a fixed cost summary. The summary format is shown below with placeholders; each run fills them with the figures it computes, or marks a field unavailable with the reason.

Dashboard artifact ("DevOps Agent Cost Dashboard"), refreshed in place each run: a totals panel, direct usage cost per day, priced downstream cost by category, tool usage by category.

Cost summary:

📊 DEVOPS AGENT COST SUMMARY
═══════════════════════════════════════════════════════════════
Billing Period:   <START ISO> → <END ISO>
Agent Space:      <SPACE_NAME> (<REGION>)
Account:          <ACCOUNT_ID>
Role Scope:       exact role only
Data Source:      GetMetricStatistics · AWS/AIDevOps

── DIRECT USAGE · space-scoped ────────────────────────────
Investigations:                 X.XX hrs              $XX.XX
Evaluations:                    X.XX hrs              $XX.XX
On-demand (Chat):               X.XX hrs              $XX.XX
Release Readiness:              X.XX hrs              $XX.XX
                                          ──────────────────
Direct Cost (before account-level credits):          $XX.XX

── CREDITS · account-level reference only ─────────────────
Support Plan:     <plan | unknown>
Credit Rate:      <XX% | unavailable>
Net Direct Cost:  unavailable at this scope

── DOWNSTREAM · exact-role scoped ─────────────────────────
Logs Insights:        X queries · X.XX GB             $XX.XX
Athena:             X executions · X.XX TB            $XX.XX
GetMetricData:          X calls · X metrics           $XX.XX
X-Ray retrieved:        X calls · X traces            $XX.XX
                                          ──────────────────
Priced Downstream Total:                              $XX.XX
                                  <exact | lower bound>

── TOOL USAGE · exact-role scoped ─────────────────────────
AWS tools:        XXX calls
Azure/3P MCP:      XX calls
Platform:          XXX calls

Scopes differ; no combined total is reported.
Paid calls:       0
═══════════════════════════════════════════════════════════════

How it differs from asking the skill in chat¶

	Skill in chat	This custom agent
Output	A plain answer	A fixed dashboard layout plus a fixed summary
Execution	Conversational	Silent: no narration, one final message
Cadence	On demand	On demand or on a schedule

The calculations are identical. The agent adds configuration, a fixed window and a consistent layout.

Setting it up¶

1. Install the skill first¶

This agent calls the devops-agent-cost-insights skill by name, so install the skill before creating the agent. Follow the skill README.

2. Create the agent¶

Operator Web App → Agents → Create agent. You can supply the agent's instructions two ways:

Import from GitHub after cloning. Clone this repository and import the agent through the GitHub integration, pointing it at this folder so SYSTEM_PROMPT.md is picked up as the instructions.
Paste the prompt directly. Copy SYSTEM_PROMPT.md from this folder and paste it into the agent's instructions.

3. Fill in the Config table¶

The prompt opens with a Config table. This is the only place these values appear; the rest of the prompt refers to them by name. Replace each placeholder:

Key	What to put	Where to find it
`ACCOUNT_ID`	The account being measured	The account the Agent Space runs in
`REGION`	The Agent Space deployment Region	`get_agent_space` / `list_associations`
`SPACE_NAME`	Display name for the report header	Your Agent Space
`SPACE_ID`	The `AgentSpaceUUID`	`get_agent_space`
`ROLE_NAME`	The role to measure, short name	The DevOps Agent role for this space
`ROLE_ARN`	The full ARN of that role	`list_associations`
`WINDOW_DAYS`	Reporting window length in days	Your choice; defaults to `15`
`SUPPORT_PLAN`	Your AWS Support plan, e.g. `Enterprise Support`. Leave blank to omit credits.	Billing console, Support plans
`PRIOR_MONTH_SUPPORT_CHARGE`	Last month's `AWS Support` charge, e.g. `5000`. Leave blank to omit credits.	Previous month's bill

To change the reporting period later, edit WINDOW_DAYS alone. The window is a trailing range computed at run time: <END> is the run time and <START> is WINDOW_DAYS before it, so a scheduled run always covers the most recent window.

SUPPORT_PLAN and PRIOR_MONTH_SUPPORT_CHARGE are optional. When set, the agent derives the credit rate from the plan and shows the plan, rate, and monthly credit as account-level reference. This agent reports one role in one space, so the credit is never subtracted here and Net Direct Cost stays unavailable at this scope; leave both blank to omit the credit lines entirely.

4. Enable CloudTrail capture for the downstream breakdown¶

The Athena identity join reads StartQueryExecution, a management event CloudTrail records by default, so grant cloudtrail:LookupEvents and confirm no tool policy blocks it. Without it, the Athena line is reported unavailable, never zero. GetMetricData and the X-Ray operations reach CloudTrail only as data events, which are off by default and add a recording charge. Enable them if you want those rows priced: the data event carries both the call count and the billable quantity. Without them the rows report as unavailable.

5. Set the schedule¶

Attach the agent to a schedule for recurring reports, or run it on demand. A weekly cadence keeps the run's own agent time small relative to the figures it reports; a daily cadence on a lightly used space can become a meaningful fraction of what you are measuring. Match WINDOW_DAYS to the cadence you choose so consecutive reports tile the period rather than overlapping heavily.