- Home
- Skills
- Testing & Debugging
- LLM Retry & Duplicate Cost Evidence Pack
More screenshots
Works with the AI Agents you already use
LLM Retry & Duplicate Cost Evidence Pack
Turn supplied LLM attempt logs into retry-chain evidence for duplicate successes, payload drift, retry storms, weak backoff and ambiguous completion.
$7
LLM Retry & Duplicate Cost Evidence Pack
Example session with this skill installed
Run the included sample attempt log through the local retry evidence pack using the supplied policy. Show duplicate successes, backoff problems and ambiguous retries. Do not send or retry any API request.
- Read your context and instructions
- Compiled the llm retry
- Generated the data export
LLM Retry & Duplicate Cost Evidence — included sample
Status
DISCREPANCY. The supplied log contains 4 attempts grouped into 2 explicit operations. There is 1 discrepancy finding, 2 manual-review findings and USD 0.100000 of observed extra-success cost from the supplied values.
| Operation | Attempt chain | Finding | Review |
|---|---|---|---|
| invoice-001 | timeout → success | BACKOFF_VIOLATION: 1000 ms before a-2 versus 2000 ms minimum | Retry after timeout lacks a stable idempotency key |
| invoice-002 | success → success | MULTIPLE_SUCCESSES: 2 successes | Supplied extra-success cost USD 0.100000 |
The generated packet includes result.json, operation-matrix.csv, anomalies.jsonl, manual-review.jsonl, RETRY_EVIDENCE.md and a SHA-256 receipt. Before rerunning work, review invoice-001’s key and backoff policy and investigate invoice-002’s two successful attempts. The tool groups only by explicit operation_id and does not contact the provider, retry anything or determine a billing fault.
operation-matrix.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
This downloadable local package turns authorized JSON or JSONL attempt evidence into a fixed, reproducible retry review packet. It includes an installable Python CLI, a local stdio MCP adapter and a companion Agent Skill. One run creates result.json, operation-matrix.csv, anomalies.jsonl, manual-review.jsonl, RETRY_EVIDENCE.md and a SHA-256 receipt. It groups attempts only by explicit operation_id, identifies measurable duplicate-success costs, weak backoff and ambiguous retry chains, and gives operators a reviewable handoff. The deterministic core rejects unsafe paths and malformed JSON, never overwrites an existing packet, escapes formula-active CSV cells and binds inputs and outputs with SHA-256. It runs locally with no API key, login, upload, telemetry or live provider action. Useful for developers investigating extra LLM spend after timeouts, automation agencies reviewing idempotency before reruns and teams documenting incidents. Requires Windows 11, Python 3.11–3.14 and buyer-supplied files in documented schemas; optional local MCP adapter uses mcp==2.2.0. Limitations: the tool never sends or retries a request. Duplicate-success cost uses supplied values; it does not establish provider billing fault.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Testing & Debugging.
The tool never sends or retries requests. Operations are grouped only by explicit operation_id. Duplicate-success cost uses supplied values and does not prove provider fault. Requires buyer-supplied local files in documented schemas.
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
Runs a local Python CLI or optional stdio MCP adapter; reads buyer-selected attempt logs and policy files; writes a new evidence packet. No browser, network or environment-variable access is required.