The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines specific shell command templates for path resolution and file operations. It mandates the use of a fallback mechanism ${CODEX_HOME:-$HOME/.codex} to ensure a base path is always present before attempting file access.
[INDIRECT_PROMPT_INJECTION]: The skill uses unvalidated input from prompts to construct file system paths, creating a surface for path traversal attacks.
Ingestion points: The <automation_id> is extracted from user prompts and interpolated directly into the AUTO_MEM path variable in SKILL.md.
Boundary markers: None. The instructions do not direct the agent to validate or sanitize the provided ID.
Capability inventory: The skill uses sed to read files, mkdir -p to create directories, and printf to modify files.
Sanitization: Absent. The lack of validation for the automation_id could allow a malicious prompt to provide an ID containing directory traversal sequences (e.g., ../), potentially leading to unauthorized access to files within the user's home directory.

automation-memory