Agent Skills: Kanchi Dividend Review Monitor

Kanchi Dividend Review Monitor

Overview

Detect abnormal dividend-risk signals and route them into a human review queue. Treat automation as anomaly detection, not automated trade execution.

When to Use

Use this skill when the user needs:

• Daily/weekly/quarterly anomaly detection for dividend holdings.
• Forced review queueing for T1-T5 risk triggers.
• 8-K/governance keyword scans tied to portfolio tickers.
• Deterministic OK/WARN/REVIEW output before manual decision making.

Prerequisites

Provide normalized input JSON that follows:

• references/input-schema.md

If upstream data is unavailable, provide at least:

• ticker
• instrument_type
• dividend.latest_regular
• dividend.prior_regular

Non-Negotiable Rule

Never auto-sell based only on machine triggers. Always create WARN or REVIEW evidence for human confirmation first.

State Machine

• OK: no action.
• WARN: add to next check cycle and pause optional adds.
• REVIEW: immediate human review ticket + pause adds.

Use references/trigger-matrix.md for trigger thresholds and actions.

Monitoring Cadence

• Daily:
  • T1 dividend cut/suspension.
  • T4 SEC filing keyword scan (8-K oriented).
• Weekly:
  • T3 proxy credit stress checks.
• Quarterly:
  • T2 coverage deterioration and T5 structural decline scoring.

Workflow

1) Normalize input dataset

Collect per ticker fields in one JSON document:

• Dividend points (latest regular, prior regular, missing/zero flag).
• Coverage fields (FCF or FFO or NII, dividends paid, ratio history).
• Balance-sheet trend fields (net debt, interest coverage, buybacks/dividends).
• Filing text snippets (especially recent 8-K or equivalent alert text).
• Operations trend fields (revenue CAGR, margin trend, guidance trend).

Use references/input-schema.md for field definitions and sample payload.

2) Run the rule engine

Run:

python3 skills/kanchi-dividend-review-monitor/scripts/build_review_queue.py \
  --input /path/to/monitor_input.json \
  --output-dir reports/

The script maps each ticker to OK/WARN/REVIEW based on T1-T5. Output files are saved to the specified directory with dated filenames (e.g., review_queue_20260227.json and .md).

3) Prioritize and deduplicate

If multiple triggers fire:

• Keep all findings for audit trail.
• Escalate final state to highest severity only.
• Store trigger reasons as single-line evidence.

4) Generate human review tickets

For each REVIEW ticker, include:

• Trigger IDs and evidence.
• Suspected failure mode.
• Required manual checks for next decision.

Use references/review-ticket-template.md output format.

SEC Filing Guardrail

When implementing live SEC fetchers:

• Include a compliant User-Agent string (name + email).
• Use caching and throttling.
• Respect SEC fair-access guidance.

Output Contract

Always return:

1. Queue JSON with summary counts and ticker-level findings.
2. Markdown dashboard for quick triage.
3. List of immediate REVIEW tickets.

Multi-Skill Handoff

• Consume ticker universe and baseline assumptions from kanchi-dividend-sop.
• Feed REVIEW results back to kanchi-dividend-sop for re-underwriting and position-size review.
• Share account-type context with kanchi-dividend-us-tax-accounting when risk events imply account relocation decisions.

Resources

• scripts/build_review_queue.py: local rule engine for T1-T5.
• scripts/tests/test_build_review_queue.py: unit tests for T1-T5 and report rendering.
• references/trigger-matrix.md: trigger definitions, cadence, and actions.
• references/input-schema.md: normalized input schema and sample JSON.
• references/review-ticket-template.md: standardized manual-review ticket layout.