The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/generate-digest.sh is vulnerable to command injection because it uses an unquoted heredoc (cat > "$DIGEST_FILE" << EOF) for file generation. This construct allows the shell to expand variables and execute command substitutions (e.g., $(...)) present in the expanded text. Since variables like $SECTIONS are derived directly from user-controlled log files, an attacker can trigger arbitrary command execution by including shell metacharacters in a log entry. There is no sanitization or escaping performed on these inputs.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted content from daily logs. Ingestion points: The scripts/generate-digest.sh script reads data from $HOME/clawd/memory/. Boundary markers: No delimiters or safety instructions are used to isolate log content from the system environment. Capability inventory: The script possesses file system write access and is vulnerable to the previously described command execution flaw. Sanitization: The extraction process does not validate or sanitize log content, allowing malicious instructions or code to pass into the generation phase.

memory-curator