- Home
- Skills
- Legal & Compliance
- bear-model-doc-writer
Works with the AI Agents you already use
Bear Model Doc Writer
Evidence-linked AML/CTF and fraud transaction monitoring model documentation from SQL, dbt and specs.
Free
bear-model-doc-writer
Example session with this skill installed
Document scenario SYN-FANIN-01 from this specification: count distinct sending account IDs per receiving account over the preceding 30 minutes, settled SEK transactions only, deduplicated by transaction ID, indicator at 5 or more senders (illustrative threshold). No implementation or executed tests are available.
- Read your context and instructions
- Compiled the bear-model-doc-writer
SYN-FANIN-01 — documentation draft
Status: Draft for review. Evidence date: 16 November 2026. Source version: fictional user specification; no code commit supplied.
1. Purpose and scope
Document a hypothetical fan-in indicator for a receiving account. The supplied specification counts distinct sending account IDs, not people, customers or devices. A match does not establish criminal intent. Product, jurisdiction and production population are unknown.
2. Timing and operational role
This describes post-transaction detection, preceding any alert review. The preceding 30 minutes are specified; exact endpoints, decision trigger, event/processing time and data availability at decision time are not. No authority to block payments is evidenced.
3. Inputs and lineage
Required by the description: transaction ID, sending account ID, receiving account ID, currency, settlement state and a transaction timestamp. Actual table/column names, join cardinality and null/error behavior are unknown. These are conceptual fields, not an invented production schema.
4. Intended logic and parameters
| Item | Supplied intent | Evidence |
|---|---|---|
| Entity | Receiving account | E1 |
| Measure | Count distinct sending account IDs | E1 |
| Window | Preceding 30 minutes; endpoints unresolved | E1 |
| Filters | Settled SEK transactions only | E1 |
| Duplicate handling | Deduplicate by transaction ID; conflicting-duplicate tie-break unresolved | E1 |
| Indicator | Count >=5; explicitly illustrative | E1 |
5. Intended versus implemented behavior
Implementation was not supplied. All logic above describes intended behavior only. There is no evidence that a deployed query uses these filters, groups or comparisons.
6. Outputs and explainability
The output schema is unspecified. Proposed review fields: receiving account, decision time, distinct sender count and contributing transaction IDs. These are suggestions, not existing reason codes or approved fields.
7. Testing and performance
Executed tests: none supplied or run. Proposed tests: 4/5/6 distinct senders; repeated transfers from one sender; exact and adjacent window boundaries; duplicate IDs; unsettled/non-SEK input; missing identifiers; data unavailable at decision time. No labels or denominators support precision, recall or false-positive rates.
8. Legitimate lookalikes and limitations
Collections for a shared expense can resemble fan-in. The specification contains no intent/context feature that would distinguish this from illicit activity. Account-level distinctness does not establish customer-level distinctness. Cross-bank visibility and reversals are unspecified.
9. Operations and governance
No owner, approval, monitoring, suppression policy or change record was supplied. The illustrative threshold is not a calibrated or approved production value.
10. Gaps and next actions
| Gap | Impact | Priority | Needed evidence / proposed reviewer |
|---|---|---|---|
| Window/time/availability semantics | Different input sets and alerts | P0 before conformance testing | Owner specification / rule owner |
| Missing implementation and lineage | Cannot verify behavior or joins | P0 before conformance conclusion | Versioned code and schema / engineer |
| Duplicate tie-break, reversals and nulls | Potential inconsistent counts | P1 | Explicit policies / rule owner |
| No executed tests or labeled evaluation | No effectiveness conclusion | P1 | Test outputs and approved evaluation data / model specialist |
11. Evidence register
E1: user-provided fictional SYN-FANIN-01 specification, example input above. Supports intent only. No code, executed results, employer data or external regulatory claim was used.
Review checklist and next actions
Intent, implementation and execution remain separate. All numerical parameters trace to E1 and remain illustrative. No approvals are inferred.
- Resolve time, duplicate and reversal policies.
- Supply versioned implementation and approved lineage if implementation review is wanted.
- Design and execute discriminating tests through an authorized local adapter; keep model-effectiveness evaluation separate.
Connects securely to your tools. The creator never sees your data.
About this skill
Turns SQL, dbt models, specifications and test results into a reviewable, evidence-linked draft of your AML/CTF or fraud transaction monitoring model or scenario documentation. It separates design intent from implemented behavior and executed tests, flags gaps with priorities, and never invents thresholds, approvals or compliance claims. Built with real-time and P2P payments in mind. Works with synthetic data by default. Not legal advice.
Continue from documentation to test design
The complete synthetic example on this page shows what the free workflow produces when only a specification is supplied. No implementation or test results are invented. If you also need test contracts, synthetic fixtures and explicit expectations, see TM Scenario Test Designer, USD 12 one-time. The free documentation skill remains usable on its own; optional research in the test designer is charged separately. Its portable file/ZIP update 1.1.0 is approved and live.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Legal & Compliance.
Does not execute supplied code or access production systems. Does not calculate precision, recall or false-positive rates without the needed labels and counts. Does not make legal or regulatory compliance conclusions, and does not certify or validate models.
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
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed today
- Free with an Agensi account
Requires
Nothing to install.
Permissions
No special permissions declared or detected