The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bash and curl to perform its primary function of sending HTTP POST requests to the ServerChan API.
Evidence: The scripts/send.sh script and the instructions in SKILL.md utilize curl -sS -X POST to transmit data to ftqq.com and ft07.com domains.
[DYNAMIC_EXECUTION]: The skill employs a Python subprocess to ensure reliable and safe JSON construction for complex notification payloads.
Evidence: scripts/send.sh uses a Python heredoc (`python3
<<'PY'`) to process notification options and escape strings, avoiding common shell interpolation vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it ingests untrusted text (like deployment logs or task titles) to send as notifications.
Ingestion points: TITLE and DESP arguments in scripts/send.sh.
Boundary markers: The skill instructs the agent to treat these as text content for a message.
Capability inventory: Capabilities are limited to curl network requests to specific vendor domains.
Sanitization: Payloads are sanitized via json.dumps in the Python-based JSON assembler to prevent schema confusion.
[CREDENTIALS_UNSAFE]: The skill manages secrets safely by requiring the use of environment variables rather than hardcoded keys.
Evidence: It utilizes SERVERCHAN_SENDKEY and includes code to verify its presence without printing the value. The use of curl -sS prevents the key (which is part of the URL path in the ServerChan API) from being displayed in progress bars or verbose logs.

serverchan