Add AI Cost and Energy Analyst agent

[pich]

Adds agents/ai-cost-and-energy-analyst.agent.md, an agent for answering "what will this AI work cost, and is it worth it" with arithmetic instead of an estimate.

It wires the ai-economics MCP server by Michał Piszczek (piszczek.pl/tools) as a local stdio server over npx, following the same pattern as comet-opik.agent.md. Twelve deterministic calculators: token cost across GPT/Claude/Gemini/DeepSeek with prompt-cache discount, context-window sizing, fully-loaded agent-hour cost including human review, model-routing savings, LLM energy and CO₂, joules per verified task, and the reviewer capacity an agent fleet actually needs.

Why this is not something the model already does well. These are multi-step numeric questions — unit conversions, per-vendor price tables, compounding rates — and they end up in budgets. A language model estimating them in-context is exactly the failure mode the agent is written to prevent: the instructions repeatedly tell it to call a tool rather than compute, and to quote the returned formula so the number can be challenged. It also pushes back where naive cost reasoning goes wrong, e.g. cheaper-per-token models frequently losing on cost per verified task because of a lower pass rate.

Notes

• No API key, no account, no secrets in the frontmatter — the server is keyless and stores nothing. (Per discussion Guidance for submissions involving paid services #968: this is a free service, not a paid one.)
• MIT-licensed; also in the official MCP Registry as pl.piszczek/ai-economics and in docker/mcp-registry (PR pending).
• docs/README.agents.md was regenerated with node ./eng/update-readme.mjs rather than hand-edited.
• The twelve tool names in the agent instructions were checked against the server's actual tools/list output.

[github-actions]

🔴 Contributor Reputation Check: HIGH risk

Check	Risk
Profile	HIGH
Credential audit	NONE

Maintainers: please review this contributor before merging.
See the workflow run for full details.
Automated check powered by AGT.

[github-actions]

🔴 Contributor Reputation Check: HIGH risk

Check	Risk
Profile	HIGH
Credential audit	NONE

Maintainers: please review this contributor before merging.
See the workflow run for full details.
Automated check powered by AGT.

[github-actions]

🔒 PR Risk Scan Results

Scanned 1 changed file(s).

Severity	Count
🔴 High	0
🟠 Medium	1
ℹ️ Info	0

Severity	Rule	File	Line	Match
🟠	package-exec-command	agents/ai-cost-and-energy-analyst.agent.md	8	command: 'npx'

This is an automated soft-gate report. Findings indicate review targets and do not block merge by themselves.

[github-actions]

🔍 Vally Lint Results

⚠️ Warnings or advisories found

Scope	Checked
Skills	0
Agents	1
Total	1

Severity	Count
❌ Errors	0
⚠️ Warnings	0
ℹ️ Advisories	1

Summary

Level	Finding
ℹ️	Vally currently lints SKILL.md content. Agent files were detected but skipped:

Full linter output

### Agent files detected (not linted by vally)
ℹ️ Vally currently lints SKILL.md content. Agent files were detected but skipped:
agents/ai-cost-and-energy-analyst.agent.md

[pich]

Context for whoever reviews the automated reputation flag on this PR, since I would rather explain it than have you guess.

The flag is awesome_fork_burst — 15 awesome-list forks in 72 hours. That is accurate, and the reason is mundane: I open-sourced a set of calculators and an MCP server over the past two weeks and have been submitting them to the directories where they belong, one fork per list. It is a burst of submissions from one person over a short window, not a network of accounts. Every PR is a single entry, written to each list's own format, and links to the same public repo.

Who I am: Michał Piszczek, https://piszczek.pl. GitHub account is 15 years old. The work here is ai-economics-mcp — MIT, published on npm as @michalpiszczek/ai-economics-mcp and in the official MCP Registry as pl.piszczek/ai-economics under a domain-verified namespace, which required proving control of piszczek.pl.

If the burst is itself the concern regardless of content, I am happy for this PR to sit until the pattern ages out — the agent is not time-sensitive. And if the agent is simply not a fit for this repo, say so and I will close it; I would rather take a clear no than leave a flagged PR in your queue.