price-action

name: price-action description: Use when the user asks for a price-action read, structural setup, bias, or any setups present on a symbol right now. Returns 0-3 ranked candidates plus structure (pivots, S/R, FVG, OB, liquidity). Read-only.

Price Action Read

Hybrid classical + ICT structural reader. Returns full structure (pivots, S/R, FVGs, OBs, liquidity pools, EMA stack, regime per TF) plus 0–3 ranked setup candidates. The skill never executes a trade; it hands off to pre-trade-checklist and position-sizer when the user wants to act on a candidate.

Prerequisites & first-run setup

Before the first invocation, verify:

1. trading-agent-skills-price-action CLI is on PATH. Test with trading-agent-skills-price-action --help. If not found: "Install the Python package — from the agent-trading-skills repo, run pip install -e . in a venv your harness can see."
2. mt5-mcp server is connected. Verify with mcp__mt5-mcp__ping. The skill needs get_rates (multi-timeframe bars), get_quote, and get_symbols (tick_size + digits). Without it, the skill cannot run.

No Calix, no news API keys, no config.toml. The CLI is pure-function — input goes in, structure + setups come out.

If (1) or (2) fail, walk the user through the fix before fanning out the rates calls.

When to invoke

• "what's the setup on XAUUSD"
• "is there a long setup on NAS100"
• "give me a price action read on UKOIL"
• "show me the structure on EURUSD"
• "any swing setup on USOIL today"
• "bias on XAGUSD"

Don't invoke for: pure quote/spread questions (use mt5-market-data), risk status (use daily-risk-guardian), session news (use session-news-brief).

Inputs (collect from user if implied)

1. Symbol — required.
2. Mode — intraday (H1/M15/M5) or swing (D1/H4/H1). Default to swing if user phrasing is ambiguous; ask if they say "today" or "this week" without other clues.
3. Timeframes override — optional list ["H4", "H1", "M15"] to override the mode default.

Workflow

1. Resolve the timeframe stack

2. Fan out MCP tools (parallel)

For each timeframe in the stack:

• mcp__mt5-mcp__get_rates(symbol=<sym>, timeframe=<tf>, count=200)

Plus once per call:

• mcp__mt5-mcp__get_quote(symbol=<sym>) — current bid/ask
• mcp__mt5-mcp__get_symbols(...) to obtain tick_size and digits for the symbol

3. Build the bundle and pipe to the skill

echo '{
  "symbol": "XAUUSD",
  "mode": "swing",
  "timeframes": ["D1", "H4", "H1"],
  "as_of": "<now in ISO 8601>",
  "current_quote": <get_quote output>,
  "symbol_meta": {
    "tick_size": "<from get_symbols>",
    "digits": <from get_symbols>,
    "contract_size": "<from get_symbols>",
    "trade_mode": "<from get_symbols>"
  },
  "rates": {
    "D1": <get_rates D1 output>,
    "H4": <get_rates H4 output>,
    "H1": <get_rates H1 output>
  }
}' | trading-agent-skills-price-action

4. Read the JSON output

The result includes:

• regime per TF and mtf_alignment (aligned_long, aligned_short, mixed, conflicted)
• Full structure (pivots, S/R levels, FVGs, OBs, liquidity pools, EMA stack)
• setups — 0..3 ranked candidates, each with id, type, side, entry_zone, suggested_entry, invalidation, stop_distance, confluence, structural_score, narrative_hint
• recent_bars_window — last 20 bars on the two lowest TFs
• warnings — sparse_bars_<TF>, mtf_conflict, no_clean_setup, etc.

5. Pick a setup and write the rationale

If setups is non-empty, examine each candidate's confluence, structural_score, and narrative_hint:

• Pick the rank-1 setup unless its narrative conflicts with a fact you can see (e.g. an obvious news event mid-bar).
• Write a selection_rationale string — one or two sentences explaining why this setup over the others.
• Set selected_setup_id to the chosen id (or "stand_aside" if you decide to skip even though candidates exist).

If setups is empty (or warnings contains no_clean_setup):

• Set selected_setup_id to "stand_aside" and selection_rationale accordingly.
• Render the structure narrative anyway (where price is relative to S/R, unfilled FVGs, etc.) — this is still useful context.

6. Narrate

Render a concise summary covering:

1. Regime + MTF alignment — one line.
2. Key levels — nearest S/R, unfilled FVGs, unswept liquidity.
3. Selected setup — type, side, entry, invalidation, narrative.
4. Hand-off offer — "Want me to run pre-trade-checklist + position-sizer for this entry?"

7. (Optional) Hand-off to checklist + sizer

If the user accepts the hand-off:

• For pre-trade-checklist: pass {symbol, side, entry: suggested_entry, stop: invalidation}.
• For position-sizer: pass {symbol, side, stop_distance} (plus the user's account/quote bundle). The setup's stop_distance field maps directly to the sizer's request — no manual tick-size conversion needed.

Health & degradation

• Missing bars on a TF → warnings: ["missing_bars_<TF>"]. Skill still runs on the available TFs.
• <60 bars on a TF → warnings: ["sparse_bars_<TF>"]. Detectors needing longer lookback skip silently.
• Conflicting MTF → mtf_conflict warning + scores penalised.
• No tradeable setup → no_clean_setup warning + empty setups. Narrative becomes "stand aside".

Common pitfalls

• Don't pipe stale bars. If as_of is significantly older than the most recent bar's time, the skill has no way to detect that — make sure your fan-out happens just before the pipe.
• Don't pre-filter the candidates. Pass the full output to the user; the LLM's job is to choose, not to censor.
• Don't combine modes. Pick swing or intraday; if the user wants both, run twice.