More screenshots
Works with the AI Agents you already use
AI Batch Billing Evidence Pack
Reconcile one OpenAI Batch run by custom_id and generate a local, redacted support packet with conservative retry candidates.
$5
AI Batch Billing Evidence Pack
Example session with this skill installed
Reconcile this completed OpenAI Batch incident from my authorized local files. The input contains invoice-001, invoice-002 and invoice-003. The output file has successful 200 responses for invoice-001 with 40 input and 12 output tokens, and invoice-002 with 42 input and 14 output tokens. The error file reports request_timeout for invoice-003. The Batch object says total 3, completed 2, failed 1, input tokens 82 and output tokens 26. My Batch-scoped dashboard observation says 4 requests, 122 input tokens and 38 output tokens. Build the packet locally and do not submit a retry.
- Read your context and instructions
- Compiled the ai batch billing
- Generated the data export
AI Batch Billing Evidence Pack
Status: DISCREPANCY
Batch status: completed
This packet compares only the supplied files and observations. A discrepancy is not proof of a billing error.
Reconciliation summary
| Measure | Count |
|---|---|
| submitted | 3 |
| delivered | 2 |
| failed_retryable | 1 |
| manual_review | 0 |
| missing | 0 |
| pending | 0 |
| duplicate_or_conflict | 0 |
| unknown_returned | 0 |
Delivered token evidence
| Measure | Value |
|---|---|
| Input tokens found in delivered rows | 82 |
| Output tokens found in delivered rows | 26 |
| Delivered rows without complete usage | 0 |
Supplied measurement comparisons
| Source | Measure | Measured | Supplied | Delta | Outcome |
|---|---|---|---|---|---|
| Batch object | request total | 3 | 3 | 0 | MATCH |
| Batch object | completed | 2 | 2 | 0 | MATCH |
| Batch object | failed | 1 | 1 | 0 | MATCH |
| Dashboard | request count | 3 | 4 | +1 | MISMATCH |
| Dashboard | input tokens | 82 | 122 | +40 | MISMATCH |
| Dashboard | output tokens | 26 | 38 | +12 | MISMATCH |
Findings
- The three dashboard measurements differ from the delivered local evidence.
- invoice-003 is a retry-review candidate because the terminal Batch returned request_timeout.
- No retry has been submitted.
Files created
result.json, reconciliation.csv, retry-eligible.jsonl, manual-review.jsonl, SUPPORT_EVIDENCE.md and evidence-receipt.json.
Safe next step
Verify that the dashboard filter covers only this Batch. Review authorization, cost and idempotency before using retry-eligible.jsonl. Send the summary and original platform exports to support only if you are authorized to disclose them.
reconciliation.csv
CSV · data export
Example file from a real run - the skill writes it into your workspace.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
When an OpenAI Batch run looks incomplete or unexpectedly billed, the evidence is split across the original requests JSONL, output JSONL, error JSONL, Batch object and dashboard observations. Manual matching is slow and makes duplicate retries more likely.
What it does
This local package matches every submitted custom_id and creates one reproducible incident packet. It detects delivered, failed, missing, pending, duplicate and conflicting records; sums delivered token usage; and compares those totals with any supplied Batch object or batch-scoped dashboard observation.
One run creates:
- result.json — authoritative structured result
- reconciliation.csv — one row per submitted custom_id
- retry-eligible.jsonl — exact original requests selected only by conservative terminal-failure rules
- manual-review.jsonl — identifiers and bounded reasons without prompt or response bodies
- SUPPORT_EVIDENCE.md — support-ready human summary
- evidence-receipt.json — SHA-256 bindings for inputs and outputs
The download includes an installable Python CLI, a local stdio MCP adapter and a companion Skill. The CLI and MCP use the same fixed reconciliation engine.
Privacy and safety
Processing stays local. The package does not read an API key, connect to an OpenAI account, submit retries, open support cases, upload files or decide whether a charge is valid. DISCREPANCY means supplied measurements differ; it is not proof of overbilling. retry-eligible.jsonl may contain original request bodies and must be treated as sensitive.
Best for
- AI freelancers preparing evidence before contacting support
- developers reconciling one failed or partial Batch run
- small teams that need a repeatable incident record before paying for a rerun
Requirements
Windows 11, Python 3.11–3.14 and authorized local OpenAI Batch JSON/JSONL files. MCP is optional and pins mcp==2.2.0.
Known limitations
V1 supports Responses, Chat Completions, Embeddings, Completions and Moderations Batch envelopes. It does not fetch account data, current model prices or dashboard data. It does not determine fault, certify a refund or automatically resubmit work.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.
V1 supports Batch envelopes for Responses, Chat Completions, Embeddings, Completions and Moderations. It requires authorized local exports and does not fetch account data, dashboard data or current model prices. It does not determine fault, certify refunds, submit retries or monitor continuously. Period-wide dashboard observations are recorded but not compared with one Batch. retry-eligible.jsonl contains selected original requests and must be reviewed as sensitive data.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 months ago
- Software
Python
Permissions
File Scopes
Local-only tool. Reads authorized relative JSON/JSONL paths below the current working directory and writes a new fixed evidence-packet directory. Rejects absolute paths, parent escapes, symlinks/reparse points and existing output directories. No network or environment-variable access.