InvoicePaymentChaser — README

Polite Overdue Invoice Follow-Up Generator
3 emails + SMS/DM + phone script + internal memo, tailored to your client
relationship and tone. No legal advice.

Quick start

1. Load the system prompt into a host agent

Open system_prompt.md and paste its full contents into the system prompt
slot of a host agent (Claude, GPT, or any Capafy-compatible runner).

2. Send an input

Either provide the 10-field list described in system_prompt.md § INPUT
FORMAT, or send a JSON object matching the schema in
sample/input/sample_invoice_input.json.

Required fields:

• invoice_amount
• payment_due_date
• days_overdue
• client_relationship — one of: new client, repeat client,
  long-term client, strategic client, difficult client, other
• tone — one of: soft, standard, firm
• payment_method

Optional fields:

• invoice_number
• client_name
• business_name
• prior_followup_history

3. Receive the follow-up package

The agent returns a Markdown document with these sections in order:

• A. Situation Diagnosis (Light Delay / Watch Closely / Firm Follow-up
  Recommended) — status + 2–3 bullets + recommended next step.
• B. Email 1: Soft Reminder — subject + body.
• C. Email 2: Deadline-Based Follow-up — subject + body + requested
  response/payment date.
• D. Email 3: Final Confirmation + Next Action Notice — subject + body, no
  legal threats.
• E. SMS/DM Short Version — 1–2 short messages.
• F. Phone Script — opening / confirmation question / payment timing question
  / calm close.
• G. Internal Record Memo — date / amount / due date / days overdue /
  contact attempt / next scheduled follow-up date.

A reference output is available at
sample/output/expected_invoice_output.md so you can see the expected shape.

4. (Optional) Pre-flight checks with the helper script

The helper at scripts/invoice_payment_chaser.py is pure Python standard
library. It does not call any LLM. Use it to validate input JSON or render an
empty output skeleton.

# Validate an input file
python3 scripts/invoice_payment_chaser.py \
    --input sample/input/sample_invoice_input.json \
    --mode validate

# Render an empty output skeleton (useful for manual editing)
python3 scripts/invoice_payment_chaser.py \
    --mode template \
    --output ./out/invoice_followup_skeleton.md

Files

Design principles

• Tactical, not theoretical. Emails are ready to copy-paste into Gmail
  and send. SMS lines fit in a single text. The phone script is short enough
  to memorize before dialing.
• Three tones, three escalation steps. Soft Reminder → Deadline-Based
  Follow-up → Final Confirmation, matched to the diagnosed status (Light
  Delay / Watch Closely / Firm Follow-up Recommended).
• Relationship-safe by default. Even at the firm tone, the language stays
  professional and non-threatening. Aggressive requests get softened with a
  brief note that the wording has been kept relationship-safe.
• Hard ban on legal language. No threats, no penalty claims, no legal
  enforcement wording, no debt collection terminology. If asked, the agent
  declines and suggests consulting a qualified professional.
• No invented invoice details. Missing required information is asked back
  or marked with [Client Name], [Invoice Number], [Payment Link]
  placeholders.

Compatibility

• Python 3.10 or newer (for the helper script)
• Any LLM host that accepts a system prompt and a user message in plain text
• Capafy listing fields (the parameter mapping in agent-card.md is aligned)

Customization

• Edit config/output_template.md to change section order or add fields
  (e.g., a localized header for your business name, or an extra
  "next-action owner" column in the internal memo).
• Add new tone descriptors to system_prompt.md for industry-specific
  styles (e.g., long-term retainer clients, public-sector buyers).
• The helper script takes a --config argument if you want to keep multiple
  template variants (e.g., one per business unit).

License

MIT. See LICENSE.