The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to clone an external repository from https://github.com/neteroster/CodexCont.git and execute its contents using the uv package manager (uv sync and uv run). This execution of third-party code from an unverified source poses a risk of arbitrary code execution.
[EXTERNAL_DOWNLOADS]: The installation process requires downloading the project source code and its dependencies from external repositories and package registries.
[COMMAND_EXECUTION]: Setup and operation involve several shell commands, including git clone, cp config.example.toml config.toml, and uv run python run.py.
[DATA_EXFILTRATION]: As a proxy for API requests, the middleware handles sensitive Authorization headers and access tokens. The inject and passthrough_then_inject authentication modes, combined with the ability to override the upstream URL via the Responses-API-Base header, create a potential surface for credential redirection, although the documentation mentions built-in safeguards to prevent leaks in specific modes.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by processing and folding untrusted data streams from upstream APIs.
Ingestion points: Upstream API responses via upstream.url and Responses-API-Base overrides.
Boundary markers: Standard SSE (Server-Sent Events) formatting.
Capability inventory: Python script execution (run.py), local server hosting, and network requests via httpx.
Sanitization: JSON parsing and event folding logic implemented in the middleware core.

codexcont-middleware