Skip to content
Back to search

AgentMeasure

SkillWorks with CodexGitHub

Inspect Codex / Claude logs, verify metering samples and re-count supported billing exports. Free local tools; bounded $10 diagnosis and $20 QA.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today

Check it deeper on your machine

It has changed since its scan on 11 Nov 2026.

As a Claude Code plugin, install it and then ask Claude Code:

/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/roy-tong/Verovo with scanskillsafety

Or from a terminal, clone it and run it:

git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/roy-tong/Verovo'

It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.

Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/roy-tong--agentmeasure/badge.svg)](/listings/roy-tong--agentmeasure)

HTML

<a href="/listings/roy-tong--agentmeasure"><img src="/api/listings/roy-tong--agentmeasure/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 215 lines

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.

  1. # Verovo
  2. <img src="assets/verovo-logo-card.png" alt="Verovo" width="300">
  3. **Know what an AI task really costs, and what its result is worth.**
  4. Verovo connects buyer logs, supplier charges, financial records and business outcomes. Review charges and delivery, allocate direct and shared costs to tasks, and calculate cost per verified success and ROI when the evidence supports it.
  5. For teams buying or operating AI services, the output is a reproducible economic record: task costs, allocation rules, verified results and explicit evidence gaps. The local `economics` command runs on an explicit case, and `intake` assembles that case from a local billing-period bundle; automated optimization and insurance remain future work. [Feature status and build/integrate decisions](docs/feature-status.md).
  6. <p align="center">
  7. <img src="assets/launch/hero.svg" alt="Verovo connects charging rules, usage and payment records, and delivery evidence into one local buyer review that reports charges and delivery as separate PASS / FAIL / UNPROVABLE verdicts" width="720">
  8. </p>
  9. [Website](https://roy-tong.github.io/Verovo/) · [中文](README.zh-CN.md) ·
  10. [Product definition](docs/product-definition.md) · [Buyer task guide](docs/buyer-review.md)
  11. ## Where this came from
  12. This repository grew out of an audit: ~110 AI usage/cost tools were fed a
  13. synthetic fixture whose expected total can be computed by hand, and the tools'
  14. own parsers were run against it. 65+ evidence-backed findings; most tools were
  15. clean, and the ones that were not failed in five specific classes
  16. (per-block re-summation, re-emitted events, cache-pricing semantics,
  17. price-table drift, resume/fork loss).
  18. - **The report:** [The Token-Accounting Bug Report](campaigns/audit-report-2026-09.md)
  19. - **Three single-page cases** with every number sourced and dated: [launch case cards](docs/launch/README.md)
  20. - **What the audit taught about its own method:** one assertion was corrected by real data, and that rule now applies to every finding ([codeburn #1264](https://github.com/getagentseal/codeburn/pull/1264))
  21. Merge counts are reported as of the date they were taken and are never kept as
  22. a single hand-edited number: **14** merged at report publication (2026-09-09),
  23. **23** merged or accepted per the claims index (2026-10-11). A GitHub search for
  24. the author's merged PRs counts the whole account, not audit fixes, so it is not
  25. a substitute for either figure.
  26. ## Three workflows for AI service buyers
  27. | Workflow | Question | Useful next step |
  28. |---|---|---|
  29. | **Charge review** | Do the charges match the applicable rules and records? | Reconcile, pay, request evidence or ask the supplier to review a difference |
  30. | **Task economics** | What does each task cost, including shared expenses? | Compare cost per verified success and evidence-backed business value |
  31. | **Delivery review** | Does the work meet the buyer’s stated requirements? | Accept, request remediation, adjust scope or decide on renewal |
  32. Run either workflow independently or combine them. Correct charges do not establish
  33. successful delivery; poor delivery does not automatically establish incorrect billing.
  34. Missing evidence remains **UNPROVABLE**.
  35. Our users are business and engineering owners accountable for AI spending or results,
  36. with finance and procurement as collaborators. AI support is an initial application
  37. scenario. Coding agents, automation platforms, tools/APIs and model services belong to
  38. the wider scope; coverage is verified task by task.
  39. ## Try task economics locally
  40. ```bash
  41. git clone https://github.com/roy-tong/Verovo.git
  42. cd Verovo
  43. python3 healthcheck/verovo economics --case examples/task-economics.json --json economics.json --markdown economics.md
  44. ```
  45. The synthetic case has two tasks, USD 43.50 in total costs and one verified success.
  46. A duplicate financial row is flagged and excluded. The verified USD 100 value yields
  47. ROI 1.2989. These are demonstration inputs, not customer results. Use your own explicit
  48. records and review coverage before making a financial decision.
  49. ## Try a buyer review locally
  50. The `review` command is part of the **v0.13.0** release.
  51. Python 3.9+, standard library only:
  52. ```bash
  53. git clone https://github.com/roy-tong/verovo.git
  54. cd verovo
  55. python3 healthcheck/verovo review --sample --html buyer-review.html --json buyer-review.json
  56. ```
  57. The synthetic example returns **charges: PASS**, **delivery: FAIL**, and one missing
  58. delivery observation as **UNPROVABLE**. Open `buyer-review.html` to inspect both lanes,
  59. criteria, rule references, observed values and evidence hashes. Every report carries
  60. three **conclusion levels**, kept separate: Level 1 rule consistency (what the run
  61. establishes), Level 2 evidence credibility (rule applicability, source truth, coverage —
  62. reported as not established when absent), Level 3 the buyer business decision (outside
  63. the tool). The six operating stages — Intake → Source Check → Rule Resolution →
  64. Reconciliation → Delivery Assessment → Buyer Decision — map each step to a command in
  65. [docs/review-method/buyer-review-stages.md](docs/review-method/buyer-review-stages.md).
  66. [View the synthetic report](https://roy-tong.github.io/Verovo/examples/buyer-review.html).
  67. Agent hosts can call the same review read-only via MCP (`verovo-mcp`: `verovo_review`,
  68. `verovo_review_template`, `verovo_recount`, `verovo_invoice_intake`, `verovo_period_intake`;
  69. no write or submit actions exposed).
  70. To review your own task, copy the [task and observation examples](healthcheck/verovo_healthcheck/demo/buyer-review.json)
  71. and follow the [input guide](docs/buyer-review.md). Checks compare supplied criteria
  72. with values in local JSON files. A person must confirm rule applicability and source
  73. truth; the tool does not infer contract terms or judge arbitrary AI output quality.
  74. ## Re-count a bill in 60 seconds
  75. Install the current CLI release:
  76. ```bash
  77. python3 -m pip install "verovo==0.13.0"
  78. verovo recount --sample
  79. ```
  80. The `verovo` project is published on PyPI. The Provider SDK
  81. (`@verovo/mcp`) has its own release number and is not tied to the CLI version.
  82. `prepare` maps supported exports; `recount` applies a selected vendor rule set;
  83. `dispute` organizes supported differences; `recovery` records supplier responses
  84. and realized corrections separately. A rule-priced finding is not recovered money.
  85. [Recount examples](examples/showcase/README.md) · [Periodic reconciliation](RECONCILE.md)
  86. ## What is available today
  87. | Capability | Current boundary |
  88. |---|---|
  89. | Buyer review (main preview) | Explicit charge/delivery criteria, local JSON evidence, separate results and offline HTML/JSON; no universal contract parser |
  90. | Supported billing recounts | Selected export formats and versioned rules; applicability must be reviewed |
  91. | Buyer Ledger | Local tool-bill processing and 26 catalog entries; cataloged does not mean connected or validated on real customer bills |
  92. | Developer diagnostics | Local coding-agent log checks, measurement conformance and regression fixtures |
  93. | Task economics (main preview) | Local task/run/usage/cost/outcome case → direct and allocated cost, cost per verified success, ROI when verified value exists; unallocated and unverified items stay explicit. Cost drivers and scenario-based proposals build on the same case |
  94. | Human buyer review | Scope and fees discussed for a concrete task; commercial acceptance and repeat purchase remain under validation |
  95. ### Task economics (main preview)
  96. `verovo economics --case case.json --json out.json --markdown out.md` connects
  97. business tasks, agent runs, financialized usage, invoices/cost items and verified
  98. outcomes. Costs are never added across currencies; duplicate financial rows are
  99. reported rather than summed; missing task/run links and shared costs without an
  100. allocation rule stay unallocated; unverified outcomes do not enter the
  101. cost-per-verified-success denominator. See the
  102. [case schema](healthcheck/schemas/task-economics.schema.json) and
  103. [design boundary](docs/task-economics.md). The command is local and makes no
  104. network calls.
  105. To avoid hand-writing the case, assemble it from a period bundle:
  106. ```bash
  107. python3 healthcheck/verovo intake --bundle examples/period-bundle --out ./period-out
  108. python3 healthcheck/verovo economics --case ./period-out/case.json --markdown economics.md
  109. ```
  110. The bundle manifest names local traces, usage rows, bills, outcomes and optional
  111. invoices. Intake reconciles a bill with financialized usage by `charge_id`, keeps
  112. unallocated costs, currency mismatches and unverified outcomes as named open
  113. items, and refuses to fabricate a precise ROI when the period is incomplete. See
  114. [period intake](docs/period-intake.md).
  115. Once the case exists, look at where the money goes and what could change:
  116. ```bash
  117. python3 healthcheck/verovo drivers --case ./period-out/case.json
  118. python3 healthcheck/verovo optimize --case ./period-out/case.json --scenario scenarios.json
  119. ```
  120. `drivers` separates recorded from attributable cost and keeps `not-recorded` /
  121. `not-separable` buckets visible. `optimize` applies explicit assumptions
  122. (`scale_cost`, `drop`) to a copy of the case, re-runs the same economics and
  123. attaches a verification plan to every proposal; nothing is auto-applied and no
  124. saving is claimed. See [cost drivers and scenarios](docs/cost-optimization.md).
  125. [Buyer Ledger](buyer-ledger/README.md) · [Coverage limits](buyer-ledger/REVIEW-20261029.md) ·
  126. [Developer CLI](healthcheck/README.md) · [Conformance](conformance/pack/README.md)
  127. ## Start with a real task
  128. [Describe a billing, acceptance or renewal task](https://roy-tong.github.io/Verovo/first-look.html#contact):
  129. name the service, account or task, decision date, available record types and owner.
  130. You can request delivery review without first finding a billing discrepancy.
  131. Free measurement lane / 免费测量通道: bring 20–100 sanitized rows to
  132. [issue #6](https://github.com/roy-tong/verovo/issues/6) or the
  133. [website form](https://roy-tong.github.io/Verovo/services.html) —
  134. 免费:带 20–100 行脱敏数据提交到 issue #6 或官网表单。
  135. Providers reviewing upstream spend are buyers in that transaction. Provider-side
  136. work for downstream billing is an extension triggered by a concrete customer dispute,
  137. revenue discrepancy or procurement requirement. It does not redefine the buyer focus.
  138. The existing [free measurement runs](https://roy-tong.github.io/Verovo/services.html)
  139. retain their separate sample-based scope. No fixed enterprise price or recovery
  140. promise is published. Scope and any fee are agreed before commissioned work.
  141. ## Evidence and contribution
  142. Local CLI commands do not upload data. Optional GitHub-hosted workflows upload to
  143. GitHub and require a private repository; do not put private bills in public issues.
  144. Synthetic fixtures and test results establish engineering behavior, not customer
  145. adoption, economic recovery or paid demand.
  146. Launch case cards — every number sourced and dated:
  147. [tokendash #37](docs/launch/case-tokendash-37.md) (one message billed as three),
  148. [claude-hud #758](docs/launch/case-claude-hud-758.md) (1-hour cache writes at the
  149. 5-minute rate), [VibeMirror-coder-001](docs/launch/case-vibemirror-coder-001.md)
  150. (an independent re-measurement through the free lane). Index and walkthrough
  151. images: [docs/launch](docs/launch/README.md).
  152. [CI](https://github.com/roy-tong/verovo/actions/workflows/conformance.yml) ·
  153. [Claims and reproductions](https://roy-tong.github.io/Verovo/claims.html) ·
  154. [Roadmap](ROADMAP.md) · [Contributing](CONTRIBUTING.md) · [Agent instructions](AGENTS.md)
  155. The standard, SDK, rules and reusable checks support these workflows and remain
  156. open source. [MIT license](LICENSE).
  157. ## Extract a supplier invoice into a reviewable draft
  158. Verovo provides an **optional, local invoice2data adapter** for supplier PDF/image/text invoice files. Extraction is not booked to the economics ledger, and it never infers taxes, credits, refunds, foreign exchange or task allocation.
  159. ```bash
  160. pip install invoice2data
  161. python3 healthcheck/verovo invoice-intake --file ./invoice.pdf --json ./invoice-draft.json
  162. ```
  163. A vendor-specific template may be needed. The result includes a source SHA-256 and explicit \`review_status: unreviewed\` and \`cost_ledger_status: not_imported\` fields. Review the original invoice and contractual rules before assigning costs. Core Verovo remains standard-library-only. See [feature status](docs/feature-status.md).

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.