The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands including grep, sed, sort, and tail to maintain an internal observation log. These commands are employed to verify observation numbering and resolve potential write-collisions in environments with parallel sessions.
[EXTERNAL_DOWNLOADS]: The skill references its own documentation and user guides hosted on GitHub (github.com/rebelytics). It is designed to fetch these files to ensure the user has access to the most current methodology and instructions.
[PROMPT_INJECTION]: The skill processes 'handoff documents' and session history to extract improvement opportunities. This ingestion of potentially untrusted data constitutes a surface for indirect prompt injection. However, the skill implements a multi-step analysis protocol to mitigate the risk of the agent obeying instructions embedded within the processed data.
[DATA_EXFILTRATION]: The skill requires persistent storage access to maintain an observation log and a cross-cutting principles file. It stages updated skill files in a 'skill-updates' directory within the user's workspace for review and manual installation, which is a core functional requirement of the meta-skill.

task-observer