The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill attempts to download and install an external AI agent skill using npx skills add https://github.com/latent-spaces/brag --skill brag -g. The source repository is not a known trusted organization.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to process video, manage files, and gather project metadata:
npx skills add ... and npx hyperframes doctor for environment setup.
ffprobe and ffmpeg for video validation and re-encoding.
git shortlog -sn --no-merges to extract contributor information.
Compress-Archive (PowerShell) or zip (POSIX) to package the final deliverables.
[DATA_EXFILTRATION]: The skill collects and exposes metadata about the repository's contributors by extracting real names from Git logs (git shortlog -sn). It also instructs the user to manually send the resulting archive to the author's contact points.
[PERSISTENCE]: The skill uses the -g (global) flag when installing the brag skill, which modifies the global agent environment across sessions.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted data to generate project descriptions:
Ingestion points: Reads the project's README and source code (SKILL.md).
Boundary markers: None identified; there are no instructions for the agent to ignore embedded commands in the source files.
Capability inventory: The agent has access to shell execution (npx, ffmpeg, git, zip) which could be abused if malicious instructions are parsed from the project files.
Sanitization: No sanitization or validation of the ingested file content is performed before generating the YAML metadata.

spark-gen