The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface. It retrieves note content from an external GitHub repository and instructs the agent to synthesize summaries ('meta-notes') from this data. Maliciously crafted content within those notes could influence the agent's reasoning or behavior during the synthesis process.
Ingestion points: Note content and indices (backlinks, clusters, tensions) retrieved from the GitHub repository via the gh CLI.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are used when interpolating note content into synthesis tasks.
Capability inventory: The skill allows network writes via curl POST requests to the SlipBox API and repository interactions via the gh CLI.
Sanitization: No input validation or sanitization is performed on the retrieved Markdown/YAML content before processing.
[CREDENTIALS_UNSAFE]: The 'Setup Check' section includes a diagnostic command that prints the first six characters of the SLIPBOX_API_KEY to the standard output. While helpful for debugging, this leads to partial credential exposure in the agent's interaction logs.
[COMMAND_EXECUTION]: The skill utilizes shell commands (curl, gh, grep, sed) to perform its operations. While these are documented and aligned with the skill's purpose, they represent a significant capability for file and network interaction.
[EXTERNAL_DOWNLOADS]: The skill interacts with external services, specifically a Vercel-hosted API (slip-box-rho.vercel.app) and a GitHub repository (Randroids-Dojo/PrivateBox). These interactions involve downloading note content and uploading synthesized data. Both platforms are well-known services.

slipbox