The Agent Skills Directory

[COMMAND_EXECUTION]: The push() function in references/push-implementation.md uses the source command to read the credential file at $HOME/.config/pushover/claude-code. This executes the file's content as shell code, posing a risk of arbitrary command execution if the file is compromised.
[DATA_EXFILTRATION]: The skill accesses a sensitive local file path at $HOME/.config/pushover/claude-code to retrieve Pushover API credentials. It also performs network communication with api.pushover.net to send data that may contain sensitive terminal output or buffer contents.
[PROMPT_INJECTION]: The skill processes untrusted data from shell arguments, stdin, and tmux buffers, presenting an attack surface for indirect prompt injection. (1) Ingestion points: references/push-implementation.md (reads from command arguments, stdin, and tmux buffers via tmux save-buffer). (2) Boundary markers: None identified in the prompt interpolation. (3) Capability inventory: references/push-implementation.md (network access via curl). (4) Sanitization: Uses --data-urlencode for HTTP parameters, but does not sanitize the transmitted message content itself.

sharing-text