| name | trading-agents-research |
| description | AUTO-INVOKE when user says ta [ticker], research [ticker], or 研究 [ticker]. Runs the real TradingAgents multi-agent framework (analysts + debate + risk + PM) for a single stock ticker. |
| references | ["scripts/call_trading_agents.py"] |
Required credentials — scripts resolve these as: env var → ~/.aisa/credentials (KEY=VALUE lines):
AISA_API_KEY — AISA multi-model gateway — default LLM + aisa-* skills (search / marketpulse / prediction-markets / twitter). No key yet? Get one at https://console.aisa.one/get-started (takes about a minute).
If a value is missing, do NOT fail silently and do NOT just tell the user to "export" it.
Guide them through setup right here in the conversation:
- If they don't have the key, give them the sign-up link above and wait.
- Ask them to paste the key into the chat, then save it for them:
mkdir -p ~/.aisa && printf 'NAME=value\n' >> ~/.aisa/credentials && chmod 600 ~/.aisa/credentials
(replace NAME=value with the real pair; keep one KEY=VALUE per line).
- Immediately retry the user's original command — the file takes effect at once, no host restart needed.
If the user prefers not to paste secrets into chat, offer the alternative: they export the
env var themselves in the host's environment, then restart the host. 不要静默失败 / never fail silently.
Trading Agents Single Ticker Deep Research
Calls the real TradingAgents multi-agent framework. Runtime ≈ 15-20 minutes.
The script runs in the background and caches the finished report on disk
(~/.tradingagents/results/<TICKER>/<date>-report.txt). Your jobs are to
(1) confirm the research started, (2) retrieve the cached report with
--resend when the user asks for the result (or after ~20 minutes), and
(3) handle any error status.
HOW TO LAUNCH RESEARCH
Step 1 — Extract the ticker from the user's message (uppercase, strip whitespace). Tell the user:
Starting TradingAgents research for <TICKER>. This takes 15-20 minutes — ask me for the result later (e.g. "resend <TICKER>") and I will fetch the finished report.
Step 2 — Call execute_code ONCE with timeout=60. Replace <TICKER> with the uppercase ticker string (e.g. "NVDA"):
import os, importlib.util, sys
script_path = os.path.expanduser("${CLAUDE_PLUGIN_ROOT}/skills/trading-agents-research/scripts/call_trading_agents.py")
spec = importlib.util.spec_from_file_location("call_ta_module", script_path)
mod = importlib.util.module_from_spec(spec)
sys.modules["call_ta_module"] = mod
spec.loader.exec_module(mod)
status = mod.run_in_background(<TICKER>)
print(status)
Step 3 — MANDATORY STATUS VALIDATION:
The output MUST start with exactly one of: STARTED:, DONE:, or FAILED:.
STARTED: → relay to user verbatim. Research is running in background; the report is retrievable later (see below). No further action needed now.DONE: → the full report follows the status line — deliver it (see delivery rules below).FAILED: → relay verbatim so user sees the error. Do NOT retry automatically.- Empty or none of the above → reply ONLY with:
ERROR: TradingAgents failed to start (status-line validation did not pass).
HOW TO RETRIEVE THE FINISHED REPORT
When the user asks for the result ("resend [ticker]" / "重新发送 [ticker]" / "研究结果好了吗"), fetch it from cache without re-running TA:
import os, importlib.util, sys
script_path = os.path.expanduser("${CLAUDE_PLUGIN_ROOT}/skills/trading-agents-research/scripts/call_trading_agents.py")
spec = importlib.util.spec_from_file_location("call_ta_module", script_path)
mod = importlib.util.module_from_spec(spec)
sys.modules["call_ta_module"] = mod
spec.loader.exec_module(mod)
status = mod.run_and_report(<TICKER>, resend=True)
print(status)
Use execute_code with timeout=120 for this call (it only reads cache, no TA run).
DONE: → the full report follows. Deliver sections 1. FINAL DECISION and 2. TRADING PLAN verbatim; then offer the remaining sections (debate, risk, analyst reports) on request or deliver them in follow-up messages. Do not rewrite or summarise the decision/plan content itself.FAILED: No cached result found → research is still running (or was never started). Tell the user to wait, or launch a fresh run if none was started today.
Parameters
ticker: Stock ticker symbol (e.g., "NVDA", "AAPL", "AMD")
Rules
- Call
run_in_background exactly ONCE per research request — never loop or retry.
- The result IS cached on disk after every successful run. Use resend to retrieve it — never a fresh run just to re-read a report.
- If you get
FAILED:, report the error to the user and stop. Do not retry automatically.
- Never compose research content from memory — 100% of user-facing research comes from the report text printed by the script.
