Track Falsifiable A-Share Theses with This Agent Skill

Where this fits

You're an analyst agent that just processed a Chinese A-share company's quarterly report. The stock price moved, but you need to know whether your original investment thesis still holds. Without a structured way to track evidence, your agent will drift into narrative re-writing or overreact to price action. This skill gives your agent a disciplined, falsifiable thesis-building and update workflow, so every research decision is auditable.

Why agents benefit

• Prevents hindsight bias: The update mode forces the agent to read the previous thesis version and append new evidence, never silently rewrite history.
• Enforces multi-perspective risk coverage: The skill mandates at least three counter-theses across categories like demand, competition, balance sheet, policy, governance, and accounting—so your agent won't fixate on a single bullish narrative.
• Separates valuation from fundamentals: The skill records valuation anchors as model snapshots, preventing price changes from being misread as evidence for or against the thesis.
• Generates actionable next steps: Output includes the strongest counter-argument and a specific next checkpoint, giving your agent a concrete research task instead of a vague "monitor" instruction.
• Provides tamper-evident persistence: The optional JSONL hash-chain script appends records under an exclusive file lock, so your agent can store thesis history without risking concurrent write corruption.

Practical scenarios

Post-earnings thesis refresh

After a company releases quarterly results, your agent invokes update. It reads the prior thesis, compares new financial metrics against each pillar's indicators, updates pillar statuses (supported, mixed, weakened, broken), and logs evidence with timestamps and sources. It then lists the strongest counter-argument and the next verification date.

Building a fresh thesis for a new coverage target

When your agent starts researching an unfamiliar A-share, it runs build to create a core thesis statement, three to five falsifiable pillars with specific metrics and failure thresholds, plus at least three failure paths. The output is a structured table that can feed downstream research automation.

Auditing an existing thesis for integrity

Before a major investment committee review, your agent runs audit to check whether the thesis was rewritten, whether only supportive evidence was recorded, or whether failure conditions were omitted. This catches drift in the agent's reasoning process and ensures the thesis remains falsifiable.

Add it to your agent workflow

Install the skill via npx:

npx skills add cyijun/china-financial-services --skill a-share-thesis-tracker

Then instruct your agent to load the skill and run the build or update mode. For example, after pulling earnings data, your agent might execute:

Use a-share-thesis-tracker with mode=update. Read the existing thesis record from evidence pack, append new quarterly data, and output the updated pillar statuses and strongest counter-argument.

The skill will enforce its hard constraints: no automatic buy/sell signals, no price-based thesis validation, and no silent overwrites.

Compare before adopting

Before committing, compare this skill against alternatives in the OpenAgentSkill registry. Look for: (1) whether the skill enforces falsifiability rather than just summarizing data; (2) how recent the maintenance is—check commits and issue responses; (3) whether the output schema matches your downstream tools; (4) if you need a different market (e.g., US equities), this skill is A-share specific but the pattern is adaptable.

Why it is worth tracking

This skill is a niche but rigorous example of structured research discipline for agents. Its 18 stars and active GitHub repo show early community interest, and the author has documented a clear schema and script for tamper-evident persistence. It's worth evaluating if you're building agents that handle fundamental equity research, especially in Chinese markets. The skill's emphasis on counter-evidence and version integrity addresses a common failure mode in LLM-based analysis. For more context, see the skill page on OpenAgentSkill. Evaluate it early if your workflow involves long-horizon investment research where memory and honesty about uncertainty matter.