A weekly review is useful when you do it regularly and the same way each time. Then you can compare one week with another.
The skill produces a report in the same format every time: totals checked against ATAS statistics, a breakdown by time and duration, and an honest assessment of whether there are enough trades to draw conclusions.
How to install
Copy the skill text and paste it into the agent chat with this line on top:
Save this as a skill named "weekly-review". If you can't save skills, follow it for the rest of our conversation.
Skill text
---
name: weekly-review
description: Review trades from the ATAS journal for a period through the ATAS MCP server: totals, breakdown by time and duration, an honest note on sample size. Use when the user asks to review their week, month or trades, or to find where they lose money ("review my week", "look at my journal", "why am I down"). Not for position sizing advice and not for orders.
compatibility: Requires the ATAS MCP server, a trade journal and the ability to run code for calculations.
metadata:
version: 0.1
updated: 2026-11-01
---
# weekly-review
Review the user's trades for a period and show what the numbers say. Reply in the user's language. Read only. Do not say how much to risk, do not recommend a system and do not place orders.
## Hard rules
- Compute every number with code from `journal_get_trades` data. Check the total against `journal_get_statistics` if it is available for this period. State any difference.
- Do not make anything up. If the journal response has no instrument, do not infer it from the price.
- Fewer than 30 trades: say so and call the findings observations. Do not write "pattern" or "you have a problem with".
- Do not judge the person. Describe the trades.
- Account data stays in this chat. Do not send it to external services or files unless asked.
## 1. Period
If no period is given, use the last 7 days. If there are no trades in them, say so and offer the last 30 days or the whole journal (its bounds are in the response: `journalFrom`, `journalTo`).
## 2. Data
`journal_get_trades` with `from`, `to` and `limit` up to 1000. If there are more records, page through them. State separately:
- records without a close time: exclude them from the calculations and say how many there are;
- accounts: how many and whether they are simulated (`accounts_list`, field `isSimulated`);
- commission: if it is zero everywhere, say that the result is before commission.
Journal times are UTC. For hours and days use the user's time zone. If you don't know it, ask or state which one you used.
## 3. What to calculate
- trades, winners and losers, net result, average win, average loss;
- best and worst day, longest losing streak;
- breakdown by entry hour, day of the week and duration;
- breakdown by side and instrument, if the data allows it;
- share of the result from the largest trade. If one trade flips the sign of the total, say so;
- different trade sizes make average win and average loss hard to compare. If sizes differ, show it.
## 4. Answer
```
Period and data: <dates, trades, accounts, what was excluded, time zone>
Totals: <table of the main numbers>
What the numbers show:
1. <observation with numbers and trade count>
2. <observation>
3. <observation>
Where the most was lost: <two groups with trade counts>
Sample size: <whether this can be relied on>
What to check next: <one specific question for the data>
```
End with one line: this is a review of past trades, trading decisions are up to the trader.
## 5. On request
Show a trade on a chart: only in a new tab, marked with `LongPosition` or `ShortPosition`. Do not change the user's charts.
How to run it
Use weekly-review. Review my journal for the last 30 days.
What you get
The agent checked its math against the ATAS statistics total, with a difference of 0, and showed the totals for 10 trades. Then come three observations: the positive result rests on one trade, position sizes differ so the averages can't be compared, and the most money was lost on trades held 15-30 minutes.
Full agent answer
Use weekly-review. Review my journal for the last 30 days.
Period and data: Oct 2, 2026, 23:02 to Nov 1, 2026, 22:02, Europe/London. Found 10 closed trades on three accounts: 1002, 1003, 1004. All are ATAS Sim simulator accounts, currency USD. There are no records without a close time, nothing was excluded.
Commission in the statistics is zero, so the results below are before commission costs. The trade records have no commission field. The trade-by-trade calculation matches journal_get_statistics: difference 0 USD.
Totals
| Metric | Value |
|---|---|
| Result | +1,125 USD |
| Winners / losers / breakeven | 1 / 9 / 0 |
| Share of winners | 10% |
| Gross profit / gross loss | +1,875 / −750 USD |
| Average winning trade | +1,875 USD |
| Average losing trade | −83.33 USD |
| Average trade result | +112.50 USD |
| Longest losing streak | 9 trades |
| Best day, Oct 9 | +1,712.50 USD, 7 trades |
| Worst day, Oct 23 | −587.50 USD, 3 trades |
What the numbers show:
-
The positive result rests on one trade. A trade on Oct 9 that lasted 7 minutes 11 seconds made +1,875 USD, 166.7% of the net result. Without it the result flips sign: −750 USD over 9 trades.
-
Position size varies a lot. The only winning trade had a size of 5, the other nine a size of 1. So the average win and average loss in money can't be compared directly. Account
1002made +1,875 USD on one trade,1003made −662.50 USD on six, and1004made −87.50 USD on three. -
In this sample the largest losses came from trades held 15-30 minutes. All three such trades were opened on Oct 23 and lost 587.50 USD. Six trades shorter than a minute lost 162.50 USD in total.
Breakdown by entry hour, Europe/London:
| Entry hour | Trades | Result, USD |
|---|---|---|
| 10:00-10:59 | 3 | −587.50 |
| 15:00-15:59 | 3 | +1,837.50 |
| 16:00-16:59 | 4 | −125.00 |
Breakdown by duration:
| Holding time | Trades | Result, USD |
|---|---|---|
| Under 1 minute | 6 | −162.50 |
| 1-5 minutes | 0 | 0 |
| 5-15 minutes | 1 | +1,875.00 |
| 15-30 minutes | 3 | −587.50 |
All 10 trades are Long, which the ATAS statistics confirm. All were opened on a Friday, so days of the week can't be compared. The journal response doesn't include the instrument, so a breakdown by instrument isn't possible.
Where the most was lost: in two non-overlapping duration groups. 15-30 minutes: −587.50 USD, 3 trades. Under 1 minute: −162.50 USD, 6 trades.
Sample size: only 10 trades over two trading days, fewer than 30. These are observations. They are not enough for firm conclusions about entry time or holding time. The result also mixes three simulation accounts.
What to check next: does the negative result of 15-30 minute holds repeat over a longer history of account 1003 at equal size?
This is a review of past trades. Trading decisions are up to the trader.
The report ends with an assessment of the sample and one question about what to check next. In the example the sample is 10 trades, and the agent said exactly that: too few for firm conclusions. You can change the request: "for the week", "only for this account", "show the worst trade on a chart in a new tab".
What you need
- An agent that can run code
Everything the agent reads becomes part of your conversation with the AI provider. More on the Security and data page.
This is a review of past trades. Trading decisions are yours.