What the Transaction Categorization Agent does
Instinct is built for standing work like this - set it once, then it runs on your word. It proposes a category for every uncoded transaction with the evidence: date, payee, amount, proposed category, evidence, review status. You approve the batch.
Categorization errors compound into wrong reports. This assistant applies your category rules consistently, records the evidence per transaction, and lists only the genuinely ambiguous ones as questions. You review the edge cases instead of the whole pile, and the same vendor codes the same way every month.
Where the Included files go in Instinct
Hand the The included files (skill.md, setup-prompt.txt, usage-prompt.txt) to Instinct in chat. It keeps the context across sessions, so the setup survives: run the setup prompt once, approve the fictional-data test, and the agent becomes a standing workflow you trigger with one line. This is the closest fit to how the agents are designed to run.
What a run looks like on Instinct
A run looks like this: you tell Instinct to run transaction-categorization with this run's inputs, and it works through the saved skill. The output - it proposes a category for every uncoded transaction with the evidence: date, payee, amount, proposed category, evidence, review status - is prepared for your review, and Instinct asks before anything is sent, spent or committed.
Setup, step by step
- Hand
skill.md,setup-prompt.txtandusage-prompt.txtto Instinct in chat and ask it to keep them for the job (for example: transaction categorization). - Run the setup prompt once; the agent runs
transaction-categorizationon fictional sample data first - approve that test before real work. - Instinct keeps the context across sessions, so the setup survives as a standing workflow.
- Trigger each real run with one line and this run's inputs; Instinct asks before it sends, spends or commits - the same gates the agent is designed around.
Honest limits on Instinct
Instinct is in early access and its capabilities are expanding; always approve the sample-data test before pointing the agent at real work.
Questions people ask
Does the Transaction Categorization Agent work on Instinct?
Yes - the agent is three plain-text files, and on Instinct they live in Instinct. The four setup steps below are the whole job: Hand skill.md, setup-prompt.txt and usage-prompt.txt to Instinct in chat and ask it to keep them for the job (for example: transaction categorization). Then run the setup prompt once; the agent runs transaction-categorization on fictional sample data first - approve that test before real work.
What does the Transaction Categorization Agent produce on Instinct?
It proposes a category for every uncoded transaction with the evidence: date, payee, amount, proposed category, evidence, review status. Instinct prepares it and asks before anything leaves your approval. Every run waits for your approval before anything is sent, spent, published or scheduled.
Is this affiliated with Instinct?
No. AI Agent Skills is an independent store. We are not affiliated with, endorsed by, or sponsored by Instinct or its makers. Platform names belong to their owners; capabilities change, so check Instinct's current documentation.
Run this agent on another assistant
Microsoft Copilot Hermes OpenClaw Muse Perplexity ChatGPT ChatGPT Dots Claude DeepSeek Town Grok Gemini
More bookkeepers agents on Instinct
Reconciliation Agent Receivables Follow Up Agent Bills Invoices Agent
AI Agent Skills is an independent store, not affiliated with Instinct or its makers. Platform names belong to their owners.