Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add michielinksee/bantou --skill correction-memorygit clone --depth 1 https://github.com/michielinksee/bantouWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/michielinksee/bantou/correction-memory)<a href="https://agentmods.dev/skills/michielinksee/bantou/correction-memory"><img src="https://agentmods.dev/badge/skills/michielinksee/bantou/correction-memory.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00029 | $0.00647 |
| Opus 5 | $0.00015 | $0.00324 |
| Sonnet 5 | $0.00006 | $0.00129 |
| Haiku 4.5 | $0.00003 | $0.00065 |
Grade A, and why
correction-memory scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 7d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 82 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Correction Memory
Persistent memory system that learns from tax accountant corrections. Every correction is permanently stored and never repeated. This is the core differentiator of the plugin — it gets smarter with every use.
When to use
Invoke this skill when:
- A tax accountant corrects a classification ("this should be 交際費, not 会議費")
- The user wants to check what past corrections exist for a vendor
- The system needs to recall learned patterns during classification
How it works
Correction flow
- Tax accountant reviews a classification result
- If incorrect, they submit a correction with:
- Original classification
- Correct classification (勘定科目 + 税区分)
- Reason for correction (optional but valuable)
- The correction is stored in local memory permanently
- All future transactions matching the same pattern use the corrected classification automatically (Memory hit in Stage 1)
Pattern matching
Corrections are stored as vendor/description patterns. When a new transaction arrives, the memory is queried for matches:
- Exact vendor name match ("スタバ 渋谷店" matches "スタバ 渋谷店")
- Normalized vendor match ("スタバ" matches any Starbucks location)
- Description keyword match (configurable per correction)
Cross-client application
Corrections apply across all client companies in the firm. If a tax accountant corrects "Zoom" from 通信費 to 支払手数料 for Client A, that correction automatically applies to Clients B, C, etc.
This is intentional — accounting conventions are firm-level decisions, not per-client. Individual client overrides are supported but rare.
Storage location
Memory is stored locally at ~/.cockpit-mcp/memory.json. This is
fully local — no data is sent to any cloud service.
The memory file is human-readable JSON and can be manually edited, backed up, or transferred between machines.
Cost reduction over time
As Memory accumulates corrections, the proportion of transactions handled by Stage 1 (free, instant) increases while Stage 2 (API call, costs tokens) decreases. Typical trajectory:
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 7d ago First seen · 82 lines · 29 tokens per session scan A d6cba21182bd
correction-memory is a skill published in the GitHub repository michielinksee/bantou (2 stars, last pushed 16d ago), licensed MIT. It adds 29 tokens to every session and 647 once invoked, about $0.0001 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
bank-reconciliation
Match book entries against bank or credit card statements to find discrepancies, missing transactions, and duplicates. Uses exact, near, and batch matching with balance verification. Trigger on "reconcile bank statement", "match transactions", "bank balance doesn't match", "close the books".
monthly-close
Repeatable month-end close checklist for small businesses — lock the period, process outstanding receipts, categorize remaining items, reconcile all accounts, review large transactions, check contractor 1099 thresholds, generate monthly P&L summary. Trigger on "close the month", "month-end close", "monthly…
depreciation-assets
Determine whether to expense or depreciate business assets using the de minimis safe harbor ($2,500), Section 179 ($1,220,000 limit), bonus depreciation (phase-out 2024–2027), or MACRS schedules. Track asset basis, useful life, and accumulated depreciation. Trigger on "depreciate", "Section 179", "bonus depreciation"…
estimated-taxes
Calculate and schedule quarterly estimated tax payments (Form 1040-ES) for self-employed individuals. Covers self-employment tax calculation (15.3%), income tax estimation, safe harbor rules (100%/110% of prior year), payment deadlines, and underpayment penalties. Trigger on "estimated taxes", "quarterly taxes"…
expense-categorization
Assign transactions to tax-aligned expense categories using vendor name, line items, and amount patterns. Handles ambiguous vendors like Amazon by inspecting line-item detail from the receipt source. Supports split categorization and learns from user corrections. Trigger on "categorize expenses", "sort transactions"…
meals-deduction
Rules for deducting business meals at 50%, required documentation (who, what, where, why), entertainment at 0% since TCJA 2018, on-premises meals going to 0% in 2026, and common mistakes. Trigger on "deduct a meal", "business lunch", "client dinner deductible", "meals deduction", "entertainment deduction", "can I…