TradingAgents

# Fix list: TradingAgents

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/tradingagents, the October 2026 research run, assessed 25 October 2026. Grade B, 64.7 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on TradingAgents: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 62 out of 100, up to 6.7 more on the total

Why it scored 62: Framework reading. 30 for what leaves the machine and how credentials are held, 20 for least privilege and approvals, 15 for prompt injection, 15 for audit and 20 for the security programme. Keys are read from the environment, and the package loads the nearest .env on import by walking up from the working directory. The ChatGPT sign-in uses PKCE and dynamic client registration, keeps tokens in a 0600 file and revokes them on logout. The only calls of its own are the interactive CLI's daily release and notice check, with no run data, off without a terminal or with TRADINGAGENTS_NO_UPDATE_CHECK=1. SEC EDGAR gets a placeholder contact address unless one is set (24 of 30). It places no orders and holds no broker credentials, the data tools only read, and every ticker is validated before it is used in a file path. No approval hooks, which nothing in it needs (14 of 20). News, Reddit and StockTwits text goes into prompts. The source names prompt injection as the reason for path validation, but no guidance on injection was found in the README (5 of 15). Saved reports record the run's settings and each agent's report, the memory log keeps every decision, and `debug=True` streams the graph (11 of 15). Security reports go to GitHub private vulnerability reporting per the issue template, and the changelog has Security sections for a langchain-core patch and a dependency removed over CVE-2026-22218. No SECURITY.md, no security.txt on tauric.ai, no bounty, and the repository's advisories page wasn't read (8 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Reliability, 67 out of 100, up to 6.6 more on the total

Why it scored 67: Read with the local-software lines, because the owner runs it. Python 3.11 or later is stated and CI tests 3.11 to 3.14, but there is no package on PyPI per shields.io. The README installs from a clone or Docker Compose, and from 0.7.0 each GitHub release carries a wheel built and smoke-tested by a workflow (12 of 20). Public CI runs 982 test functions across four Python versions and one time zone away from UTC, a clean-install import check and Ruff on every push and pull request, and shields.io's badge read passing on main on 25 October 2026 (25). 43 open and 486 closed issues per shields.io, which gives counts but not their age or labels, so crash and regression issues couldn't be separated (18 of 25). The changelog follows Keep a Changelog and Semantic Versioning and lists breaking changes under "Upgrading from", but 0.5.1, a patch release, moved import paths and removed classes (10 of 15). Version 0.7.0 with no stability statement (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 3. Schema & documentation, 61 out of 100, up to 6.3 more on the total

Why it scored 61: Framework reading. No API reference or docs site. Typed interfaces exist in the source, Pydantic models for the research manager's, trader's and portfolio manager's outputs and Annotated signatures on the twelve data tools, with no py.typed marker (10 of 25). The README and changelog are Markdown in the repository; no llms.txt (4 of 10). The README states limits plainly, that runs vary, which statements past-dated runs withhold and why, and that it is a research tool, not advice (15 of 20). Configuration is a plain dict of string keys, though environment overrides are coerced to the default's type and fail loudly on a bad value (8 of 15). Runnable Python and CLI examples throughout the README; the vendor error hierarchy is documented only in source docstrings (9 of 15). Dated changelog for every release from 0.1.0 with upgrade notes (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 4. Agent ergonomics, 61 out of 100, up to 6.3 more on the total

Why it scored 61: Framework reading, scored on what an outside agent needs to call it. No MCP client or server, so it can't be attached as tools; a run is four lines of Python (`TradingAgentsGraph(config=...).propagate(ticker, date)`) or one headless CLI command with flags (10 of 25). Analyst selection, debate and risk rounds (default 1), `max_tool_rounds` (20), `max_recur_limit` (100), article limits and `--no-html` size a run (14 of 20). A typed vendor error hierarchy separates unavailable from absent data, the headless CLI names the missing flag before a run starts, and an unparseable decision returns REVIEW rather than Hold (15 of 20). Checkpoint resume is opt-in, a backtest rerun with the same `run_id` skips finished cells, and settlement writes each decision once; nothing writes to an outside system (15 of 20). Sensible defaults, but the default provider is OpenAI's GPT-6 models with a key, and Python only (7 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 5. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: No payment protocol (0). Self-hosted rule. Free under Apache 2.0 with nothing to buy from Tauric Research to run it (20), no card (20) and no Tauric account (20). Tauric Research's hosted services at tauric.ai are invitation or enterprise access with no public price and aren't the graded surface. Model provider costs are the user's.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 6. Transparency & trust, 68 out of 100, up to 2.8 more on the total

Made of editorial 73, provenance 62.

Why it scored 68: The editorial half. Apache 2.0 (30). Local software. The README says where the memory log, checkpoints, cache, reports and ChatGPT sign-in file live, and names each data source and model provider a run calls. tauric.ai's privacy policy covers its hosted services, not the framework, and gives no retention periods (18 of 30). "Upgrading from" notes in a dated changelog list what was removed or moved, with no written deprecation policy, and 0.5.1 removed APIs without a deprecation period (8 of 20). The update check is disclosed in the README and changelog with an opt-out, and the code says what it sends; the README doesn't name api.tauric.ai (17 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Domain age: tauric.ai, registered 2025-03-26 (1 year) (3 of 15)
- Status page: not found (0 of 10)
- security.txt: not found (0 of 10)

## 7. Maintenance & community, 82 out of 100, up to 1.6 more on the total

Why it scored 82: 0.7.0 on 16 October 2026 (30). Six releases in the 90 days to 25 October, 0.4.0, 0.5.0, 0.5.1, 0.5.2, 0.6.0 and 0.7.0 (20). 486 issues closed against 43 open and 908 pull requests closed against 80 open per shields.io, and each release credits the people who reported or fixed its issues. Reply times couldn't be read, and 229 of the 236 commits in 90 days are by one maintainer's accounts (17 of 25). No PyPI package; versions ship as GitHub releases with a wheel since 0.7.0 (8 of 15). CI with tests, lint and a clean-install check; dependencies set lower bounds only and no Dependabot configuration is in the repository (7 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: github.com pages. The shell got 403 from github.com and the GitHub API, and WebFetch refused the URLs, so github.com/robots.txt, the releases page, issue reply times, Discussions and the security advisories page weren't read. Counts of issues, pull requests, stars, CI status and the latest release come from shields.io's GitHub badges; the code, tags, changelog and history from a clone through the session's git proxy.
- unchecked: whether a package named tradingagents exists on PyPI under someone else. pypi.org answered with a bot check, and shields.io reported the package or version not found. The README doesn't name PyPI.
- Star count is shields.io's rounded 114k; the exact number wasn't read.
- firstReleased is 5 June 2025, the v0.1.0 tag and the commit "initial public release of TradingAgents". The repository's first commit, 28 December 2024, held only the paper's page, figures and citations, which matches the paper's first arXiv version the same day.
- The README and tauric.ai describe the portfolio manager approving a trade that is then executed on a simulated exchange. No exchange, order or broker code was found in v0.7.0; the output is a rating and report. Recorded as a description of the paper's simulation, not deducted. Judgement call.
- The "up to 30.5% annualized returns" on tauric.ai is the paper's backtest and Tauric Research's claim. The changelog records look-ahead fixes to historical runs from 0.2.3 to 0.7.0, after the paper's figures; whether they bear on those figures wasn't established. Not deducted.
- Legal entity. The terms and privacy policy say Tauric Research, under California law, with no legal form or address.
- Category. frameworks, as the brief proposed. It is a library and CLI that runs its own fixed agent graph; agent-harnesses was considered, but it doesn't run a general agent loop for a person or edit files and run commands.
- Our own products. TradingAgents doesn't do the job of LocalGhost or letme. It analyses one instrument with a fixed team of agents and picks or routes no tools. Graded the normal way.

## Weaknesses

- No MCP client or MCP server, so an agent drives it only through Python or the CLI
- Not on PyPI; installs from a clone, Docker or the wheel attached to each GitHub release since 0.7.0
- Documentation is the README and changelog only, with no API reference or llms.txt
- 0.5.1, a patch release, moved import paths and removed classes without a deprecation period
- News, Reddit and StockTwits text goes into prompts with no documented guidance on prompt injection

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Run headless with `tradingagents --ticker <T> --date <YYYY-MM-DD> --analysts market,news,fundamentals --save --no-show`; a missing answer stops the run and names the flag
- Guard on the REVIEW signal before mapping `propagate()`'s decision to a rating; it means no rating could be parsed
- Set SEC_EDGAR_USER_AGENT to the operator's own name and address before fetching US fundamentals
- Keep provider keys out of the working directory tree unless intended, since the package loads the nearest .env on import
- Treat the rating as model output for research; the framework places no orders and its results vary between runs

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.