Investigation reporting
Skill yogeshg665/sleuth-fraud-investigator/skills/investigation-reporting
Sleuth - Agent Skills that automate payment fraud investigations, plus a deterministic Python engine as the executable reference. Explainable, deterministic scoring and decisions with tokenized card references and human-review gates.
npx -y skills add yogeshg665/sleuth-fraud-investigator --skill investigation-reportingAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Produces an audit-ready case report and a human-readable narrative summarizing the transaction, signals, score, decision, and recommended actions. WHEN: "write the case report", "summarize the investigation", "generate fraud report", "explain the decision", "audit trail", "case narrative".
SKILL.md
2.2 KB, 400 tokens by cl100k_base, as published. Nobody here has run it
Investigation Reporting
Overview
Generates the final deliverable: a structured JSON report and a narrative summary suitable for an audit trail. The narrative is produced deterministically by default; an optional language model may enrich it but must not alter the score or decision.
When to Use
- As the final phase of every investigation, after
fraud-decisioning.
Inputs
| Input | Required | Description |
|---|---|---|
case | yes | The scored case with its signals. |
decision | yes | The recommendation from fraud-decisioning. |
Process
- Assemble the structured report: case identifier, transaction summary, risk score, decision, confidence, triggered signals with rationale and evidence, and recommended actions.
- Build a deterministic narrative covering the transaction, the score, the recommendation, the contributing indicators, and the next actions.
- If a language model is configured, request an enriched narrative using the deterministic version as the grounding context. On any failure, keep the deterministic narrative.
- Attach the narrative to the decision and return the report.
Outputs
A structured report dictionary and a narrative string.
Reference Implementation
src/fraud_investigator/agents/reporting_agent.py.
Rationalizations
| Excuse | Rebuttal |
|---|---|
| "Let the model decide the wording of the verdict." | The verdict comes from scoring and policy; the model only narrates. |
| "Omit evidence to keep the report short." | Evidence is what makes the report auditable; include it. |
Red Flags
- The narrative states a different outcome than the decision.
- A model failure blocks report generation instead of falling back.
Verification
- The report contains the score, the decision, the confidence, every triggered signal, and a narrative consistent with the decision.