The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill is designed to automatically fetch data from external sources including the GitHub Search API, GitHub Releases API, and the npm Search Registry to identify new repositories and packages related to its functionality.
[REMOTE_CODE_EXECUTION]: The primary objective of the pipeline is to "queue approved improvements for implementation." This involves retrieving code, configurations, or techniques from the internet and integrating them into the local environment. This creates a high risk of executing malicious code if an attacker-controlled repository or package is approved by the automated agents.
[COMMAND_EXECUTION]: The system relies on a daily automated scan script (hooks/daily-improvement-scan.js) and provides a command interface (/improve) that facilitates the management and execution of the improvement queue, which involves file operations and potential script execution.
[PROMPT_INJECTION]: The skill exhibits a significant surface for Indirect Prompt Injection (Category 8).
Ingestion points: The skill ingests untrusted data from GitHub repository descriptions, release notes, and npm package metadata as described in SKILL.md.
Boundary markers: There are no documented delimiters or safety instructions in the pipeline's logic to prevent the "Evaluator" or "Security" agents from being influenced by malicious instructions embedded in the scanned external content.
Capability inventory: The skill and its underlying toolkit ("CC Commander") possess broad capabilities including file system access, command execution, and the ability to modify their own code/skills (evolution-engine.js).
Sanitization: The process relies on a "Security Agent" (an LLM-based check) to identify risks, which lacks the robustness of formal sanitization and is susceptible to adversarial bypass techniques.

continuous-improvement