The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell utilities like grep, sort, and tail via the command line to manage observation numbering within its persistent log.md file.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. Ingestion points: Monitors all task execution sessions and user feedback (SKILL.md). Boundary markers: Absent. Capability inventory: Shell command execution, file system read/write, and automated instruction generation for other skills (SKILL.md). Sanitization: Includes multi-layer PII stripping but lacks mechanisms to detect or filter adversarial instructions in observed data.
[SAFE]: The skill contains a 'Confidentiality Safeguards' section defining a four-layer protocol to ensure that proprietary or client-identifying information is not leaked into generated skill files.
[SAFE]: References to the author's GitHub repository for feedback and the use of 'npx skills' for installation are consistent with the vendor's own infrastructure and are documented neutrally.

task-observer