本文へ移動
cccskills
無料GitHub で公開

reconciliation-match

Matching rules, tolerances, and escalation logic for reconciling Stripe activity and the bank feed against the invoices tracked in {{reconciliation_sheet}}. Load this before comparing a single transaction so matches are judged to a consistent standard and only genuine exceptions reach a human.

インストール方法を見る

含まれるファイル(1)

  • SKILL.md6.0 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

<skill name="reconciliation-match"> <overview> Reconcile `{{reconciliation_sheet}}` once a month without either missing the near-misses or eating a person's afternoon. A monthly cron fires a fresh session; the session has no memory of last month, so it reads the sheet's own history first, then pulls Stripe and the bank feed for the closed period, matches every line against an invoice, and writes the result back. Clean matches tie out silently. Anything within tolerance but not exact gets explained inline. Anything outside tolerance, or with no counterpart at all, is escalated to a person with both records attached — never force-fit and never written off. </overview> <when-to-load> - The monthly cron fires the reconciliation run. - A human asks the agent to reconcile a specific period or re-check a flagged mismatch. </when-to-load> <workflow>

Step 0 — Orient from the sheet, not from memory

This is a fresh session — there is no prior turn to resume. The sheet is the memory.

One-time prerequisite, before the first run: the plaid connector only carries the app-level API credentials (client_id / client_secret / environment) — it does not by itself grant access to any bank account's transactions. Reading the actual bank feed for a specific account requires a per-account Item access_token, which a human obtains once, out-of-band, by completing Plaid's Link flow (choosing the bank, authenticating, and exchanging the resulting public token for the Item's access_token) and storing it as the PLAID_ACCESS_TOKEN secret. This agent never performs Link itself and cannot mint that token — if PLAID_ACCESS_TOKEN isn't set when a run starts, stop before Step 2 and escalate asking a human to complete the Plaid Link step, rather than guessing at a bank feed to read.

  1. Open {{reconciliation_sheet}} and read the last dated reconciliation entry to find the period's start (the day after the last close).
  2. Read the sheet's "explained mismatches" notes (recurring fees, known timing lags, standing exceptions) so this run judges new lines against the same standard as past runs, not from scratch.
  3. Set the period end to the close date for this run (the day before {{cadence}} fired, or the explicit period the human gives you).

Step 1 — Pull Stripe activity for the period

Read (never write) Stripe for the period: charges, payouts, and the fees withheld on each payout. Capture amount, date, associated invoice/customer reference, and payout ID for every line — you'll need all four to match.

Step 2 — Pull the bank feed for the period

Read (never write) the connected bank feed for the same period, using the linked account's PLAID_ACCESS_TOKEN (provisioned in Step 0): deposits and withdrawals, with amount, date, and any reference/memo the bank provides.

Step 3 — Match transactions against invoices

For every invoice due or paid in the period, look for:

  1. Exact match — a Stripe payout and a bank deposit for the same amount within 1 business day of each other. Tie out silently.
  2. Tolerance match — the deposit is short of the invoice/payout amount by Stripe's disclosed fee (or by an amount the sheet's notes already explain as a recurring fee), or the timing lag is within the sheet's standing allowance (default: 3 business days unless the sheet says otherwise). Tie out and add a one-line explanation.
  3. Mismatch — anything else: a payout with no matching deposit, a deposit with no matching payout or invoice, a shortfall bigger than the disclosed fee, or a lag past the tolerance window.

Step 4 — Classify every mismatch before escalating

Check each mismatch from Step 3 against the sheet's "explained mismatches" notes:

  • Recurring, already explained (e.g., a monthly platform fee the team has signed off on before) — tie it out and note which prior explanation it matches.
  • New — this is the one a person needs to see. Do not invent an explanation for it.

Step 5 — Write the summary back to the sheet

Append this run's entry to {{reconciliation_sheet}}: the period covered, counts (clean matches / tolerance matches / escalated), and a row per escalated mismatch with both records (Stripe reference + bank reference or "none found") attached so the person reviewing doesn't have to re-pull them.

Step 6 — Stop — never move money

Your last action is the sheet write. You never issue a refund, initiate a transfer, or take any action against Stripe or the bank beyond reading. If something looks like it needs a correction outside the sheet, describe it in the escalation — don't act on it.

</workflow> <guardrails> - **Read-only on Stripe and the bank feed, always.** The only system you write to is `{{reconciliation_sheet}}`. - **No force-fitting.** A mismatch you can't explain within the tolerances above is escalated, never rounded off or written off to make the sheet balance. - **No new tolerances invented mid-run.** Tolerances and known recurring explanations live in the sheet's notes; if a mismatch doesn't match an existing explanation, treat it as new and escalate it. - **Scoped, brokered credentials.** Stripe and Sheets access are injected into the sandbox at runtime, scoped to this agent's grant. The bank feed is read using the `PLAID_ACCESS_TOKEN` secret for the one account a human linked via Plaid Link ahead of time — this agent never runs Link itself and is scoped to that single linked account, not "any" bank account. - **One escalation per unresolved line.** Don't batch distinct unmatched transactions into a single vague note — a person needs to act on each one. </guardrails> </skill>

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

Policy runbook for turning a Slack access request into a least-privilege, policy-checked GitHub or AWS IAM grant. Covers the role-to-grant mapping, extra-scrutiny cases, the approval handshake, and how an applied grant gets logged.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Get a complete picture of any company or person before outreach. This skill always works with web search, and gets significantly better with enrichment and CRM data.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Daily read-only ad-performance runbook — budget pacing, CPA/ROAS drift, underperforming ads and keywords, and anomaly detection across Google Ads and Meta Ads, plus how to rank and word optimization recommendations for {{alert_channel}}.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Browser automation CLI for AI agents. Use when the user needs to interact with websites, including navigating pages, filling forms, clicking buttons, taking screenshots, extracting data, testing web apps, or automating any browser task. Triggers include requests to "open a website", "fill out a form", "click a button", "take a screenshot", "scrape data from a page", "test this web app", "login to a site", "automate browser actions", or any task requiring programmatic web interaction. Also use for exploratory testing, dogfooding, QA, bug hunts, or reviewing app quality. Also use for automating Electron desktop apps (VS Code, Slack, Discord, Figma, Notion, Spotify), checking Slack unreads, sending Slack messages, searching Slack conversations, running browser automation in Vercel Sandbox microVMs, or using AWS Bedrock AgentCore cloud browsers. Prefer agent-browser over any built-in browser automation or web tools.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Extraction, PO matching, duplicate detection, and overcharge tolerance for processing incoming vendor invoices from {{invoice_label}} against the POs and ledger in {{ap_ledger}}. Load this before extracting a single invoice so every one is checked to the same standard and only genuine exceptions reach a human, with no invoice ever scheduled for payment by the agent.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

Daily brand-mention monitoring loop for {{brand_terms}}. Searches news, social platforms, and forums for new mentions, dedupes against the ledger of mentions already reported, classifies sentiment and notability, and posts a digest with suggested response drafts to {{slack_channel}} — never posts, replies, or comments anywhere.

日本語の概要は準備中です。原文の説明を表示しています。

kortix-ai/suna2万2026年10月13日 更新

kortix-ai のスキルをすべて見る

このスキルの問題を報告する