AgentReceipt
Use this skill to log your actions during multi-step tasks. At the end, you get a shareable receipt link to hand back to the user.
How to use this
When you start a multi-step task (processing files, making API calls, anything with more than one action), follow this pattern:
1. Start a session at the beginning of the task:
node {baseDir}/agentreceipt.js start-session "<short task name>"
Save the SESSION_ID from the output. You will need it for every call after this.
2. Log each significant action as you go:
node {baseDir}/agentreceipt.js log-event <SESSION_ID> <type> "<name>" "<one-sentence description>"
Pick the right type for what you did:
| Type | When to use |
|---|---|
tool_call |
External actions: reading files, calling APIs, sending emails |
llm_call |
Model responses: parsing documents, classifying data, generating text |
decision |
Choices you made: approving a payment, routing a ticket, skipping a step |
3. End the session when the task is done:
node {baseDir}/agentreceipt.js end-session <SESSION_ID>
This marks the session as complete.
4. Share the session to get a receipt URL:
node {baseDir}/agentreceipt.js share-session <SESSION_ID>
This gives you a RECEIPT_URL. Share it with the user so they can see what you did.
Example
User asks: "Process the invoices in my downloads folder"
node {baseDir}/agentreceipt.js start-session "Process invoices"
→ SESSION_ID=abc-123
# Do the work, logging as you go
node {baseDir}/agentreceipt.js log-event abc-123 tool_call "read-files" "Read 3 invoice files from downloads"
node {baseDir}/agentreceipt.js log-event abc-123 llm_call "extract-data" "Extracted totals from each invoice"
node {baseDir}/agentreceipt.js log-event abc-123 decision "approve-total" "Total $1,240 within budget"
node {baseDir}/agentreceipt.js end-session abc-123
→ Session ended.
node {baseDir}/agentreceipt.js share-session abc-123
→ RECEIPT_URL=https://agentreceipt.co/s/xyz...
Tell the user: "Done. Here's your receipt: <RECEIPT_URL>"
Guidelines
- Only log actions that matter. Skip trivial steps like reading config or formatting output.
- Keep descriptions short. One sentence.
- Never include sensitive data in descriptions. No passwords, API keys, tokens, personal information, email addresses, full file paths, or file contents. Describe what you did, not the raw data.
- Always call end-session when the task is done, even if something went wrong along the way.
- Do not log the same action twice.