Technical TA Engine
A standardized technical-analysis layer for crypto trading that computes an indicator bundle (RSI, MACD, etc.) from OHLCV data and returns a 0–100 composite score with confidence.
The skill itself is read-only indicator analysis; no malware, credential theft or covert exfiltration observed; data sources (OHLCV, optional Nexus) are disclosed and trading interpretation is transparent. However, permission boundaries, user confirmation and rollback are not stated in SKILL.md, and side effects of the exchange fetcher dependency are undeclared. Deducted for missing least-privilege declaration, confirmation and rollback mechanisms.
SKILL.md and persona.md are broadly self-consistent (Tier-0 contract, ta_indicators bundle, sample I/O), and the ta/TA-Lib dependency is echoed in pyproject and CI. But no direct evidence of a dedicated test path for this skill, undefined failure feedback for abnormal input (insufficient bars, fetcher errors), and TA-Lib C-library installation is a common failure point with no diagnostic guidance in the skill files. Static cap 10; deducted for thin error handling and failure feedback.
Scenarios are fairly clear (crypto TA queries, backtest CoT input) with direct-query examples and a query interface. But non-fit boundaries are undeclared (insufficient data, non-crypto assets), trigger semantics are imprecise, no Chinese-language support is mentioned, and reliance on overseas LLM/Nexus services may impair mainland-China usability. Deducted for insufficient evidence on boundaries, trigger precision and environment fit.
Docs are readable, AGPL-3.0 license is genuine, and the repo shows SECURITY.md and maintenance signals. But no skill-level version, changelog, FAQ or known-limitations disclosure; TA-Lib install burden is documented in README but not in the skill docs; limited progressive layering. Deducted for incomplete skill-level versioning and limitation disclosure.
The output contract (JSON schema, example values) is well defined and could feed downstream arbitration directly if the implementation is correct. But static review cannot verify indicator computation correctness or direct usability; only two few-shot examples with minor inconsistencies (RSI=72 input vs 72.5 output); marginal value over calling a TA library directly is weakly evidenced. Static cap 7; deducted for unverified results and usability.
Static cap 5. The repo has a real CI workflow (TA-Lib install, pytest) and a test suite, indirectly supporting engineering credibility, but no dedicated test file covering this TA engine's key paths appears in the evidence; README backtest return figures (+8.15%) are self-reported, unverifiable and explicitly disclaimed as non-live-edge. Deducted for lack of skill-specific independent reproducible evidence and for marketing-style figures requiring caution.
- This is a static source review only; nothing was executed. Indicator computation correctness and runtime behavior are unverified.
- TA-Lib C library installation is the main environmental hurdle; Windows or no-sudo environments may fail, and the skill files themselves provide no troubleshooting guidance.
- The skill depends on overseas LLM APIs (OpenAI/Atlas Cloud) and optional Nexus services; mainland-China reachability is uncertain and no Chinese-language support is documented.
- README backtest return figures are self-reported with small samples and must not be treated as evidence of live trading edge.
- Publisher is unverified by the FollowSkills registry; identity is unknown (not suspicious per se), but attribution and maintenance responsibility require user assessment.
- Side effects and permission boundaries of the exchange OHLCV fetcher are not declared in the skill files; review the fetcher implementation under src/tools before deployment.
What does this skill do, and when should you use it?
The Technical TA Engine is one skill within AI Market Maker (AIMM), an open-source agentic trading system, located at the Technical Analysis desk (2.3) of the operator. It takes a ticker and OHLCV market data, computes indicators including RSI, MACD, Bollinger Bands, EMA, SMA, ATR and Ichimoku, and fuses multiple signals into a 0–100 composite score with a 0–1 confidence value. The skill ships explicit trading-interpretation rules — e.g. RSI above 70 is overbought, a lone MACD cross is weak without trend or volume confirmation — and can optionally factor in Nexus as-of context such as Fear & Greed, funding and headlines. It is queryable both in natural language and through a structured query interface, and is designed to be used inside backtest chains of thought.
The skill reads OHLCV bars for a ticker such as BTC/USDT with a configurable period and bar count, then computes the full indicator catalog (RSI, MACD, BB, EMA, SMA, ATR, Ichimoku) using the ta Python library or a custom implementation. It answers direct natural-language queries like "What's the RSI of BTC?" or "Show MACD for ETH", and also exposes a structured query: /technical_ta_engine?ticker=BTC/USDT&period=4h&bars=100. Output is the full ta_indicators bundle plus composite (0–100) and confidence (0–1). Interpretation rules include: RSI > 70 overbought / bearish tilt, RSI < 30 oversold / bullish tilt; MACD crosses need confirmation from trend (EMA 20 vs 50) or volume; 3+ aligned indicators raise composite and confidence, while a mixed tape stays near 50 with low confidence. When Nexus as-of context is present, Fear & Greed ≤ 25 is treated as risk-off (fade breakouts, prefer shorts/holds) and ≥ 75 as risk-on (do not fade every dip).
- A crypto trader who wants a quick read of current RSI or MACD for a coin plus a confidence-weighted directional tilt
- A quant researcher building backtests who needs a reproducible technical-indicator scoring component with explicit interpretation rules
- An AIMM system user who needs the Technical Analysis desk to feed indicator inputs into the Alpha synthesis layer
- A trader who wants sentiment context (Fear & Greed, funding, headlines) folded into technical signal interpretation via Nexus
- A developer who wants to batch-fetch multi-period indicator bundles programmatically through the query interface
What are this skill's strengths and limitations?
- Broad indicator coverage — RSI, MACD, Bollinger Bands, EMA, SMA, ATR and Ichimoku in a single bundle
- Explicit, auditable interpretation rules (RSI thresholds, MACD confirmation, multi-indicator alignment logic) rather than a black box
- Structured output (composite 0–100 + confidence 0–1) is easy for downstream systems and backtests to consume
- Optional Nexus sentiment context (Fear & Greed, funding, headlines) prevents purely mechanical signal reading
- Depends on external data: requires an exchange OHLCV fetcher, with optional reliance on the Nexus Skills API
- Requires the ta Python library (or custom indicator code), and the wider repo needs the TA-Lib C library — a nontrivial setup
- Example backtest results in the README (e.g. Sharpe 1.79) describe the whole system, not this skill; there is no standalone performance evidence for the skill itself
- Interpretation rules are simplified (e.g. fixed RSI 70/30 thresholds) and may not be optimal across all market regimes
How do you install this skill?
This skill ships inside the olaxbt/ai-market-maker repository (one of 14 bundled skills) and is not installed standalone. Get the collection: git clone https://github.com/olaxbt/ai-market-maker.git, then install dependencies with uv sync --extra dev (install the TA-Lib C library first — conda install -y ta-lib -c conda-forge is recommended, apt on Ubuntu, brew on macOS). The skill file lives at src/agents/operator/2.3_technical_ta_engine/SKILL.md. Where to place the skill folder for your specific Agent Skills client is not documented in the source.
How do you use this skill?
Two ways: 1) ask directly in natural language, e.g. "What's the RSI of BTC?" or "Show MACD for ETH"; 2) use the structured query interface: /technical_ta_engine?ticker=BTC/USDT&period=4h&bars=100, which returns the full indicator bundle plus status. Outputs are the ta_indicators bundle, composite (0–100) and confidence (0–1). Within a backtest chain of thought, apply its interpretation rules (RSI thresholds, MACD confirmation requirement, Fear & Greed risk regimes).