The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions to perform logging and archival operations "silently" and "without interrupting the user," which functions to conceal background activity. It also directs the agent to modify project configuration files (such as CLAUDE.md) to ensure the skill is persistently loaded at the start of every session.
[COMMAND_EXECUTION]: The skill uses shell commands like grep and sort to manage metadata in observation logs and attempts to establish persistence by registering recurring tasks via system utilities or platform-specific shortcut tools.
[EXTERNAL_DOWNLOADS]: Documentation in SOURCE.md provides a command to fetch updates from a remote GitHub repository using the GitHub CLI and decoding the response via base64. While targeting a well-known service, this allows for the remote retrieval of executable content.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting data from external "handoff documents" and session histories (SKILL.md) to influence agent behavior. It lacks explicit boundary markers or sanitization, potentially allowing embedded malicious instructions to influence the agent's logic. (Ingestion points: SKILL.md; Boundary markers: Absent; Capability inventory: bash subprocesses, file system writes, task scheduling; Sanitization: Absent)

task-observer