The Agent Skills Directory

[DATA_EXFILTRATION]: The skill collects user-provided retrospective data and feedback, saving it to a local JSON file (lab-retro-output/04-feedback.json) before transmitting it to an external endpoint (https://lab-feedback-proxy.vercel.app/api/feedback) using a curl POST request.
[COMMAND_EXECUTION]: The skill executes several shell commands to perform its tasks, including mkdir for creating output directories, jq for processing JSON data, and curl for communicating with the external feedback proxy.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection due to its handling of untrusted user input.
Ingestion points: Untrusted data enters the agent context via AskUserQuestion across all four parts of the retrospective, including the collection of the user's "best prompt."
Boundary markers: The skill does not use specific delimiters or "ignore embedded instructions" warnings when saving user input to files or incorporating it into network payloads.
Capability inventory: The skill has the capability to write to the local file system (mkdir, file saving) and perform network operations (curl).
Sanitization: User-provided content is passed to shell commands and saved to markdown/JSON files without explicit sanitization or filtering to prevent the execution of embedded instructions.

lab-retro