The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements an 'Automatic learning loop' that journalizes invocations and auto-synthesizes 'playbook candidates'. These playbooks contain shell commands with entity slot substitution ({slot}) that are subsequently executed by the agent. This dynamic generation and execution of shell commands based on a local learning database introduces a risk of command injection or the execution of unintended instructions if the learning store is manipulated.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to pull notes, quotes, and transcripts from the Snipd API into a local SQLite corpus. Since this content originates from third-party podcasts and transcripts (attacker-controlled external sources), it could contain adversarial instructions intended to influence the agent's behavior when the content is searched or synthesized.
[COMMAND_EXECUTION]: The skill relies heavily on the Bash tool to execute a variety of subcommands (pull, filter, quote, synthesize, auth, learnings, etc.). It also allows for background execution of teaching commands (e.g., teach &), which reduces visibility into the operations being performed.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of external binaries via npx (from @mvanhorn/printing-press-library) and go install (from github.com/mvanhorn/printing-press-library). These resources are hosted on public registries and GitHub repositories. While these belong to the skill's own vendor, they represent external code that is downloaded and executed on the host system.
[DATA_EXFILTRATION]: The skill includes an --deliver webhook:<url> feature that allows the output of any command to be POSTed to an arbitrary URL. If an agent is coerced into using this flag with a sensitive query, it could result in the exfiltration of podcast data or local corpus content to an external endpoint.

pp-snipd