agentleFS
MendurimMendurim/strikegrok-trading-desk/skills/desk-post-trade-review/SKILL.md

desk-post-trade-review

Skill0 starsChanged 2 months agoMITView on GitHub

How the Trade Reviewer keeps the desk journal and reviews trades from the exchange's own record - grading process separately from outcome, measuring what execution actually cost, and drawing one repeatable finding per review - plus the weekly desk review and the incident review. Use after any send, when a trade closes, on the weekly routine, or when the user asks how something went.

2 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Write it under ## review in the proposal, send the block from agents/trade-reviewer.md by DM to the Desk Lead and the user, and set status: …

    line 69
  2. Distinctive:

    No recommendations about what to trade.

    line 82

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Journal and review

The journal is what the desk remembers. The review is how it improves. Both are built from the exchange record first and the conversation second, and both keep process and outcome strictly apart.

1. The journal

One file per active day at /workspace/trading-desk/journal/YYYY-MM-DD.md, appended in time order.

14:02 UTC  SG-20260910-01  opened    ADA-USD long idea (user); status evidence
14:12 UTC  SG-20260910-01  risk      PASS 4800 ADA, stop 0.1995, 0.5% stressed (limits v3)
14:29 UTC  SG-20260910-01  approval  "approve SG-20260910-01" (user)
14:31 UTC  SG-20260910-01  sent      bracket ADA-USD long 4800 @ 0.2100 GTC, tp 0.2310, sl 0.1995
                                     cloid SG-20260910-01-entry; order id 1839201122; read back status 2
16:05 UTC  SG-20260910-01  fill      4800 @ 0.21003, maker, fee $0.29
09:12 UTC  SG-20260910-01  closed    tp at 0.23100, fee $1.44; flat; sl cancelled 09:13
10:00 UTC  limits                    v3 -> v4: position count 3 -> 4 (user; adding HYPE-USD)
11:40 UTC  incident INC-20260911-01  unknown result on SG-20260911-02; looked up by cloid at 11:41,
                                     absent; corroborated 11:46; replacement approved with a fresh id
18:00 UTC  note                      maker entries at 10 bps filled inside 2h on both attempts this week

Append only. A correction is a new line beginning correction:, never an edit to an old one. Every line carries a UTC time, and an id wherever one exists. Opinions belong in reviews, not here.

2. Reviewing a trade

Triggered when a proposal reaches closed, or whenever the user asks. List every input with its timestamp:

  • the proposal file — ticket, PASS, approval, execution, reconciliation
  • GET /v2/history/fill across the window: price, size, fee, side, maker or taker
  • GET /v2/history/order by symbol and status — 2 open, 3 filled, 4 canceled, 5 untriggered, 6 rejected, 7 expired — to establish what rested when and what was cancelled
  • GET /v2/closedPositions for the realised result
  • GET /v2/history/funding for what holding it actually cost
  • where available, the Market Analyst's depth read at send time, to compare expected slippage against realised

Then measure:

MeasureHow
Entry slippage(average fill − ticket price) / ticket price, in bps, signed against the trade
Exit slippagethe same for the exit, against its ticket or trigger price
Feesfill fees in USD and as bps of notional; note maker against taker
Fundingthe sum actually charged over the holding period
Net resultrealised PnL after fees and funding, in USD and in R, with R taken from the ticket
Protectionwas a reduce-only stop resting for the position's entire life? Any gap, in minutes
Lifecycleevery stage present, in order, with timestamps; approval by id; one send; reconciled
Holding timefirst fill to flat

Grade two things, separately:

Process — clean, minor break, or major break, naming the stage. A major break is any send without a PASS or without approval by id, any period without protection, any resend on an unknown result before looking it up, or any limit breached.

Outcome — the result in R and USD, stated without adjectives.

These are independent, and saying so out loud is the point. "Process: clean. Outcome: −0.9R at the stop" is a good trade that lost money. "Process: entry sent before the PASS. Outcome: +2R" is a bad trade that made money, and the review says so plainly.

Then one finding: a leak, a control that earned its keep, or a break — chosen because it will recur. One, not a list.

Write it under ## review in the proposal, send the block from agents/trade-reviewer.md by DM to the Desk Lead and the user, and set status: closed.

3. The weekly review

From the journal, the proposals and the exchange record for the week:

  • proposals opened, rejected, voided, executed, closed
  • trades closed: count, hit rate, average win and average loss in R, expectancy in R, largest loss, largest equity drawdown
  • costs: fees and funding in USD and as a share of gross PnL
  • process: breaks by type; incidents and where each stands
  • limits: what changed and why
  • one pattern worth the user's attention, stated as a fact pattern rather than advice

No recommendations about what to trade. Strategy questions go to the Strategist, sizing questions to the Risk Manager.

4. Incident review

For each INC-YYYYMMDD-NN: the timeline from the journal and the exchange record, what the desk did, what the controls did, the exposure carried during the incident, the root cause where it is knowable, and one corrective action with an owner and a date. Blameless in tone and exact in fact. Written to /workspace/trading-desk/journal/incidents/<id>.md and linked from that day's journal.

Pitfalls

  • Reviewing from memory of the conversation rather than from fills. The chat records what people meant; the fills record what happened.
  • Letting the outcome colour the process grade. Grade the process first, then look at the money.
  • Counterfactuals. "If we had held" is not evidence of anything.
  • Ten findings. Nobody acts on ten.
  • Editing an old journal line. Append a correction.
  • Counting only status 2 when checking whether protection existed. An untriggered stop is status 5, and missing that turns a protected position into a phantom incident.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.