Use CoTrading skills in Claude Code

This guide targets native local Claude Code on Windows 11 x64. Consult the actual skill page for the package version and host verification record. Documentation of an installation mechanism is not evidence that every package/host combination has completed a model task. Official instructions checked 2026-10-30: Claude Code skills.

Before installation

You need a working Claude Code installation and your own authorized model access. Obtain the selected host package from its skill page. Read the README, license, dependencies and permissions. Method packages are free; Claude/model and optional data charges are separate. Never paste credentials into CoTrading pages or reports.

The current 0.1.0 A-batch packages—S01 earnings review, S02 thesis check, S04 rule specification and S05 ETF verification—all require Python 3.12 for their bundled deterministic tools. Those tools use only the Python standard library: no third-party Python packages or network access are needed for the teaching inputs. The AI host may still use a remote model. Text inputs and no live market feed do not remove the Python requirement.

In PowerShell, py -3.12 --version is a read-only runtime check, not installation or task verification. If Python 3.12 is unavailable, stop the fixed-tool task and prepare it separately using the package instructions in an isolated environment. Do not silently install dependencies, elevate privileges or change global Python settings.

For S01, the ZIP must unpack to one cotrading-earnings-review folder with SKILL.md immediately inside, plus the manifest, bilingual READMEs, notices, references and examples. Check download size and SHA-256 against the selected version's page. PowerShell can inspect the actual downloaded file:

$packagePath = Read-Host 'ZIP path'
Get-FileHash -LiteralPath $packagePath -Algorithm SHA256

A matching hash establishes byte identity, not independent safety or task success.

Install only the selected folder

Personal directory: %USERPROFILE%\.claude\skills. Copy the entire selected folder there. The resulting S01 file is %USERPROFILE%\.claude\skills\cotrading-earnings-review\SKILL.md. Avoid a second ZIP/version directory level between the skill folder and SKILL.md.

If the destination exists, stop. Compare versions and retain local edits in a backup outside scanned skill directories before choosing an update. Do not silently overwrite it or leave duplicate active copies.

In the Claude Code conversation, type / and select the skill, or invoke /cotrading-earnings-review. This is conversation input, not a PowerShell command. If missing, check nesting and SKILL.md name, then use the current host's refresh behavior or reopen the session. See troubleshooting.

First task

Copy examples into a separate research folder. First ask the selected skill to read only and list method version, inputs, outputs, dependencies and gaps. Once ready, request the sample review into a new reports/{run-id}/. Do not alter originals.

For S01, verify revenue +12%, profit +10%, cash flow -8% and cash-flow/profit 69.70% in the synthetic DEMO-CONSUMER example. Management's collection explanation remains attributed and missing aging stays unknown. S02 checks the dated old thesis; S04 stops for confirmation; S05 must match the four-day example. Seeing the skill name alone is not completion.

Dependencies, updates and removal

All four A-batch methods retain the Python 3.12 requirement above, including S01/S02/S04 text tasks. S04's tool prepares a specification; it does not run a return experiment before confirmation. Scanned PDFs or other formats may need additional tools beyond the standard-library samples. Review any additional setup separately and keep it isolated. Do not run a legacy bulk installer or replace the fixed tool with model mental arithmetic when it cannot run.

Updates use an explicit selected version and retained backup. To uninstall, remove only the chosen CoTrading skill folder after identifying it; retain the separate research inputs and reports. This guide does not authorize deleting other skills or modifying global model settings.