agentleFS
NT-NinjaTraderNT-NinjaTrader/mcp-skills/skills/alerts-composer/SKILL.md

alerts-composer

Skill8 starsChanged 4 months agoApache-2.0View on GitHub

Translates natural-language alert intent into a validated Tradovate alert-DSL expression. Submits the expression via create_alert.expression. Covers 37 numeric functions across Account (netLiq, dollarOpenPL, dailyLossLimit, ...) and Contract (lastPrice, percentChange, settlementPrice, ...) entities. Also covers Position (netPos, posOpenPLUsd, posInitMarginUsd, ...) and Currency (currentRate) entities. Supports arithmetic (plus, minus, times, divide, parens), comparison (gt, gte, lt, lte, eq, neq), and top-level AND/OR/XOR logic. Validates the expression offline (subject syntax, function catalog, paren balance, flat-logic rule) before submission. Use when the user says "alert", "notify", "notify me", "watch for", "wake me when", "tell me when", "heads up when", "ping me if", "set a trigger", or "let me know if". Also use when the user asks about a sibling skill's proposed alert expression. position-watchdog, event-watch, and pretrade-risk also submit their proposed expressions through this skill.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 1 is distinctive, and 6 are 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.

  • 6

    Very specific

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

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

    Use references/pattern-library.md as a routing table.

    line 39
  2. Very specific:

    If the user's intent needs a multi-bar indicator (ATR, RSI, MA, or similar), the DSL cannot express it.

    line 41
  3. Very specific:

    Use the symbol field. contract-intel also does this end-to-end.

    line 51
  4. Very specific:

    If the expression is invalid, fix it and validate it again.

    line 73
  5. Very specific:

    Use message for the plain-English translation, so the user sees it in the alerts list.

    line 100
  6. Very specific:

    When those skills emit an expression string, the user typically says "submit it" or "go ahead." Run the expression through validate.py …

    line 116
  7. Distinctive:

    Not submitted — say the word and I'll create it."

    line 124

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 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

alerts-composer

Purpose

Turn "alert me when X" into a valid alert-DSL expression. Validate the expression offline, check for duplicates, and submit it via create_alert.expression.

The DSL has:

  • 37 numeric functions across 4 entity types (Account 14, Contract 9, Position 13, Currency 1) — see references/dsl-functions.md.
  • Arithmetic + - * / with parens for grouping.
  • Single comparison per compare-expression.
  • Flat AND / OR / XOR chain at the top level — no parenthesized sub-expressions in logic.

See references/dsl-grammar.md for the full grammar.

Environment routing

Demo (simulation) and live are two separate MCP servers. Account names are unique to one server. Once a workflow resolves an account on a server, every downstream call must go through that same server. This includes my_portfolio, market_snapshot, place_order, create_alert, history tools, etc. Cross-routing fails or hits the wrong environment.

MCP tools used

Tool names below are bare. The NinjaTrader MCP server provides them. Your client adds its own prefix. See AGENTS.md at the repo root.

  • search_contracts — resolves a bare code to a symbol (the DSL subject must be an exact contract symbol).
  • my_portfolio — gives account name context. Prune with fields=["account.name"] for just the account name.
  • list_alerts — checks for duplicates.
  • create_alert with the expression param — the submit path (Mode B).
  • market_snapshot — gives price context when you translate a "near current price" intent into an absolute level. Prune with fields=["snapshots[].symbol","snapshots[].lastPrice","snapshots[].tickSize"].

Workflow

1. Classify the intent

Use references/pattern-library.md as a routing table. It has 18 patterns that cover most cases: price cross, breakeven, P&L floor, margin, session break, divergence, FX, and more. Pick the closest template.

If the user's intent needs a multi-bar indicator (ATR, RSI, MA, or similar), the DSL cannot express it. Snapshot the value now and emit a fixed-price bracket (e.g., entry ± 2×ATR_now). Tell the user this explicitly.

2. Resolve the symbol

The subject must be an exact contract symbol: ESU6, not ES. If the user gave a bare code, resolve it first:

search_contracts(text=<code>)

Use the symbol field. contract-intel also does this end-to-end.

3. Compose the expression

Paste a template from references/pattern-library.md. Substitute the symbol, account, and numeric thresholds. Or build the expression from the grammar in references/dsl-grammar.md.

4. Validate offline

echo '<expression>' | python3 scripts/validate.py

Checks:

  • Balanced parens
  • Comparison operator present
  • No quoted subjects
  • Flat-logic rule
  • Function names in the 37-function catalog
  • Subject regex [\$@]?[\w\s\-\+/|]+

Output: {"valid": true|false, "errors": [...], "warnings": [...], "functions_used": [...], "subjects_used": [...]}.

If the expression is invalid, fix it and validate it again. Always validate before you submit. A typo still passes create_alert, because the server parser accepts it. But the alert never fires, because the function or subject does not resolve.

5. Check for duplicates — required in every mode, including drafts

list_alerts(status="Active")

This duplicate check belongs to alert composition, not to alert submission. When the user asks for a draft or a validate-only check ("don't submit it"), still run list_alerts. Report the duplicate status in the answer. A validated draft that collides with an active alert is a footgun: the user may submit it blind later. Never present a draft or submit an alert without this check.

list_alerts does not return the raw expression string. Compare the new alert against each existing alert's symbol, trigger, and price fields instead. If an existing alert matches on all three, ask the user for a decision. Offer to replace the alert or keep both. Never auto-dismiss the duplicate.

Draft-only mode stops here. Report the expression, the validation result, and the duplicate status. Do not call create_alert.

6. Submit — wait for user approval first

Show the validated expression and the exact payload below. Then stop and wait for the user's approval. Never call create_alert from this skill before the user approves that payload. After the user approves it, submit the payload:

create_alert(
  expression="<validated>",
  message="<optional human-readable>",
  validUntil=<optional ISO timestamp>,
  triggerLimits=<optional int>
)

Use message for the plain-English translation, so the user sees it in the alerts list. Use validUntil for time-of-day conditions, because the DSL cannot express them.

7. Echo back the plain-English translation

"Created alert 4321: lastPrice(ESU6) > 7200 AND posOpenPLUsd(ESU6) < -300 — fires when ES prints above 7200 and your ES position is losing more than $300. Valid until 16:00 ET."

Integration — alert-proposal handoff from sibling skills

Three skills propose alert expressions:

  • pretrade-risk — after the user approves a bracket, it proposes a breakeven-trip or target-approach alert. See ../pretrade-risk/references/risk-rules.md § alert-proposal hook.
  • event-watch — proposes a reaction-window bracket around a known volatility event.
  • position-watchdog — proposes stop-proximity or P&L-floor guards on live positions.

When those skills emit an expression string, the user typically says "submit it" or "go ahead." Run the expression through validate.py first. Then submit the expression through create_alert as Mode B. Trust but verify. Never skip validation just because another skill produced the expression.

Output idioms

  • Happy path: "I'll submit posOpenPLUsd(ESU6) < -300 — fires when the open loss on your ES position exceeds $300. Sound right?"
  • Draft-only: "Draft validated: lastPrice(ESU6) > 7600 — fires when ESU6 prints above 7600. No duplicate among your 3 active alerts. Not submitted — say the word and I'll create it."
  • Invalid intent (indicator): "The alert DSL has no RSI function. I can snapshot RSI now and emit a fixed-price alert, or you could use a session-break alert (lastPrice(ESU6) > highPrice(ESU6))."
  • Invalid intent (cross-account): "Aggregate P&L across sub-accounts can't be expressed in a single alert. Options: one alert per account, or raise/lower dailyLossLimit on each upstream via update_risk_settings."
  • Duplicate: "You already have alert 4218 with this expression. Keep both, replace, or cancel the new one?"

Disambiguation

  • vs direct create_alert Mode A (a simple price cross: symbol + trigger + price): for a literal "alert me when ES crosses 7200 up", Mode A is simpler. The server composes the DSL internally in that mode. Use this skill instead when the intent needs any arithmetic, logic, or any non-price function.
  • vs sibling-skill proposals: those skills own the intent-to-expression translation for their own domain (risk, events, watchdog). This skill owns DSL composition, validation, and submission for any ad-hoc intent that does not come from another skill.

Known gotchas

Subject has NO quotes. Despite what the create_alert MCP docstring shows, the parser rejects lastPrice("ESU6"). Write lastPrice(ESU6) instead.

No parens around logic. The parser rejects (A > B) OR (C < D). Write A > B OR C < D instead.

Explicit non-goals

  • No alert cancellation logic. dismiss_alert is a separate MCP tool. This skill composes + submits only.
  • No modification. There is no modify_alert — dismiss and recreate.
  • No timer / scheduling. The DSL has no time arithmetic. Use validUntil at submission, or gate upstream with a human workflow.
  • No indicator / historical-bar math. The catalog is spot-quote and current-state only. Snapshot a value now and emit a fixed-price bracket when the user wants an indicator-driven alert.
  • Never invent functions. Stick to the 37 in dsl-functions.md.

Resource layout

  • scripts/validate.py — offline DSL lint. Stdin or --expression. Returns JSON with errors/warnings/functions/subjects. Exit code 0 on valid, 1 on errors.
  • references/dsl-grammar.md — full grammar (subject regex, operators, precedence, JSON shorthand, runtime statuses, known gotchas). Load it when you build an expression from scratch, not from a pattern.
  • references/dsl-functions.md — the canonical 37-function catalog. It groups functions by entity (Account/Contract/Position/Currency) and includes an intent-to-function routing table. Load it when you pick a function.
  • references/pattern-library.md — 18 ready-to-paste patterns (price cross, breakeven, P&L floor, session break, divergence, spread, FX, reaction window, position flip, etc.). Load it when you classify the intent or compose the expression.

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.