The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The instinct-cli.py script allows users to import instinct configurations from arbitrary URLs via urllib.request.urlopen. These files contain instructions that the agent incorporates into its behavioral model.
[COMMAND_EXECUTION]: Various scripts execute system commands and utilities. For example, instinct-cli.py uses subprocess.run to interact with git for project detection, while the observer pipeline scripts use tail, wc, and the claude CLI to manage log processing.
[PROMPT_INJECTION]: The skill ingests tool inputs and outputs into a local log file, which is then processed by an LLM to generate 'instincts'. Because this data can originate from untrusted external sources (e.g., website content or file reads), it presents a surface for indirect prompt injection that could influence the agent's learned behaviors.
[DATA_EXFILTRATION]: Tool interaction data, which may contain sensitive context, is logged and subsequently transmitted to a remote LLM provider for pattern analysis. Although the skill implements a regular expression-based scrubber to redact common secrets like API keys and passwords, this heuristic approach cannot guarantee complete protection against sensitive data leakage.

continuous-learning-v2