The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands in its preamble and completion phases to manage telemetry data. It executes cat, grep, sed, mkdir, and curl to read configuration files and log events to a local file (~/.superstack/telemetry.jsonl) and a remote endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple local files and external sources to generate video content.
Ingestion points: Reads .superstack/idea-context.md, .superstack/build-context.md, package.json, and scans CSS/Tailwind configuration files.
Boundary markers: None identified; data is read directly using cat and grep without explicit delimiters.
Capability inventory: Executes shell commands via npx (package installation), curl (network requests), and performs filesystem operations.
Sanitization: No sanitization or validation of the ingested context files is performed before they are processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill installs external dependencies and rule modules during its setup phase.
Fetches the Remotion agent skill from the official Remotion development GitHub repository.
Installs numerous NPM packages related to the Remotion framework, including @remotion/transitions, @remotion/light-leaks, and @remotion/google-fonts.
Downloads external binaries for transcription using npx remotion add @remotion/install-whisper-cpp.
[DATA_EXFILTRATION]: The skill implements a telemetry mechanism that transmits metadata to an external server.
It collects the platform type (via uname), skill name, execution phase, and duration.
This data is sent via curl to a URL defined in ~/.superstack/config.json. The skill prompts for user consent before activating this behavior.

marketing-video