The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a 'learning loop' (via the teach, recall, and playbook commands) that ingests user-supplied queries and stores them locally to influence future agent behavior and command selection.
Ingestion points: The teach and playbook amend commands in SKILL.md ingest arbitrary strings from user queries into the tool's local state.
Boundary markers: No technical delimiters or 'ignore' instructions are described for the stored data, relying instead on manual PII stripping instructions.
Capability inventory: The skill uses nepra-pp-cli, which can execute shell commands, perform network requests, and write to local files.
Sanitization: No explicit technical sanitization or validation is described for the stored queries or the command-slot substitution mechanism.
[DYNAMIC_EXECUTION]: The skill's 'playbook' feature allows the execution of sequences of shell commands defined in JSON structures with variable substitution ({slot}). This runtime assembly and execution of commands based on stored data is a potential security risk if the local learning store is manipulated.
[DATA_EXFILTRATION]: The CLI includes a --deliver webhook:<url> feature that allows the output of any command to be sent to an arbitrary external URL via an HTTP POST request. While a documented capability, it provides a built-in mechanism for exfiltrating data.
[COMMAND_EXECUTION]: The skill provides instructions for installing and executing a custom binary, nepra-pp-cli, which handles the primary logic of the skill.
[EXTERNAL_DOWNLOADS]: The skill fetches its primary CLI tool from the author's NPM registry (@mvanhorn/printing-press-library) and GitHub repository (github.com/mvanhorn/printing-press-library). These resources are managed by the skill's author.

pp-nepra