How is this different from asking a model for a stock analysis?

A plain prompt gets fluent text with numbers of unknown origin. This skill makes the agent pull figures from primary filings (sec.py reads SEC XBRL data), grade every piece of evidence, hold a bull/bear debate before setting a rating, run two adversarial audit rounds, and register each call in a ledger with a stated falsifier and due date.

What is in the package?

The SKILL.md workflow, method files (framework, judgment, equity / industry / macro flows, indicators, pitfalls, quality bar, desk), pipeline step files, three standard-library Python tools (sec.py, ledger.py, desk.py), templates and a disclosure template. The Claude Code plugin on GitHub also adds five role agents.

Which agents does it work with?

Any agent that reads the SKILL.md standard, including Claude Code, Codex CLI, Cursor, Gemini CLI and OpenClaw. The tools need Python 3.9+ and a shell. Agents without subagents play the five roles one after another.

Does it need API keys?

No keys. The SEC asks every client to identify itself, so set SEC_USER_AGENT to your name and email. Prices for the coverage watch come from a CSV you supply.

How are calls tracked and settled?

ledger.py keeps an append-only ledger. Each call has a falsifier, a horizon and a benchmark; desk.py lists calls that are coming due or whose falsifier was hit by a new filing. You settle each one with evidence and a one-sentence lesson. The hit rate counts hits divided by hits plus misses; open calls are excluded.

TopxAI Equity Research

Example session with this skill installed

Pull the last eight quarters for NVDA from SEC filings.

Read your context and instructions
Compiled the topxai equity research

Summary

Ran scripts/sec.py facts NVDA on 2026-10-13. The raw SEC companyfacts response (4.5 MB) stayed in the local cache; the agent received 3.3 KB of JSON. Evidence grade [A]. Source: https://data.sec.gov/api/xbrl/companyfacts/CIK0001045810.json

Last eight quarters (USD bn, calendar quarters)

Quarter	Revenue	Gross profit	Operating income	Net income	Operating cash flow	Capex
CY2024Q3	35.1	26.2	21.9	19.3	17.6	0.8
CY2024Q4	39.3	28.7	24.0	22.1	16.6	1.1
CY2025Q1	44.1	26.7	21.6	18.8	27.4	1.2
CY2025Q2	46.7	33.9	28.4	26.4	15.4	1.9
CY2025Q3	57.0	41.8	36.0	31.9	23.8	1.6
CY2025Q4	68.1	51.1	44.3	43.0	36.2	1.3
CY2026Q1	81.6	61.2	53.5	58.3	50.3	1.8
CY2026Q2	96.2	72.1	63.7	59.7	24.1	2.7

Quarters come from the XBRL facts: discrete quarters where the filing reports them, otherwise year-to-date differences. For example, CY2026Q2 operating cash flow = 74.4 (six months to 2026-07-26) − 50.3 (first quarter) = 24.1.

Latest balance-sheet items

From the 10-Q for the period ended 2026-07-26, filed 2026-08-26 (USD bn):

Cash and cash equivalents: 22.4
Long-term debt: 33.4
Remaining performance obligations: 3.2
Shares outstanding: 24.1bn (as of 2026-08-21)

Next steps

Cross-check these numbers against the research-lane notes and list any conflicts in notes-crosscheck.md.
Run scripts/sec.py filings NVDA for recent 8-K items before fixing the report's basis dates.

Connects securely to your tools. The creator never sees your data.

About this skill

TopxAI Equity Research turns an AI agent into a research desk. It takes a company, a value chain or a macro question from kickoff to a finished report, then tracks every call the report made and settles each one when it comes due. It is the method behind the research published at topxea.com, released free under Apache-2.0.

What it enforces

Every key number traces to a source with a link. If it cannot be found, the report says "not verified".
Evidence grades run from [A] primary filings to [D] unconfirmed reports. No [D] input may hold up a core conclusion.
One basis date in three layers: the last closed session, the financial period end and filing date, and a filing cutoff.
Every core conclusion states what would prove it wrong.
Ratings are settleable: 12 months, against a named benchmark, from the publication-date close.

What's inside

A ten-step pipeline (kickoff, research lanes, bull/bear debate, writing, judgment, two audit rounds, publish, watch, settle, update). Each step names its output files and an exit gate.
Method files: a cause / process / outcome framework, nine-step flows for single-stock, industry and macro reports, a quality bar, a catalogue of accounting traps, and an indicator library with primary sources (SEC EDGAR, FRED, BLS, Census, FINRA).
Three Python tools that use only the standard library: sec.py (SEC XBRL facts and filings; the raw JSON stays in a local cache), ledger.py (an append-only call ledger with settlement lessons) and desk.py (the coverage watch).
Templates for the kickoff note, the source cross-check, audit decisions, final checks and disclosures.
The Claude Code plugin on GitHub adds five role agents: collector, bull, bear, auditor and translator.

Limits

The skill supplies the method, not market data; price triggers read a CSV you provide. The tools contact only sec.gov and data.sec.gov, with no telemetry and no account. A full company deep dive can use millions of tokens. Output is general research information, not personalised investment advice, and ratings are research judgments, not orders.

Source, plugin and docs: https://github.com/UnbelievableT/topxai-equity-research

How to use

Skills on Agensi run inside your AI agent. Copy the prompt, paste it into your agent, and it connects to Agensi and installs this skill for you. If a payment is needed, your agent shows it to you first.

Reviews

No reviews yet

Be one of the first to try it. Every listed skill passes our trust checks below.

Security scanned

Passed our 8-point scan before listing

Fresh listing

Recently published to Agensi

Free forever

No account required to browse

TopxAI Equity Research