The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill downloads an installation script from https://raw.githubusercontent.com/endlessblink/watchpost/main/install.sh and executes it locally using the bash interpreter. This allows for arbitrary code execution from a remote source outside the skill's bundle.
[COMMAND_EXECUTION]: Multiple shell commands are used to manage the dashboard, including curl for status checks, npm start for launching the application, and lsof/kill for process management. These commands operate with the permissions of the local environment.
[COMMAND_EXECUTION]: User-provided port values are directly interpolated into shell commands (e.g., lsof -ti:PORT). Without strict validation that the input is a number, this pattern is vulnerable to command injection where a malicious user or indirect input could append additional commands.
[EXTERNAL_DOWNLOADS]: The skill initiates network requests to download the install.sh script from GitHub during its setup phase.
[PROMPT_INJECTION]: The skill processes MASTER_PLAN.md to update its internal state, creating a surface for indirect prompt injection.
Ingestion points: The skill reads and parses docs/MASTER_PLAN.md (specifically sections like 'Roadmap' and 'Active Work').
Boundary markers: There are no defined delimiters or instructions provided to the agent to treat the markdown content as untrusted data.
Capability inventory: The skill possesses extensive capabilities, including executing shell scripts and managing local processes.
Sanitization: No sanitization logic is described to filter or escape potentially malicious instructions embedded within the markdown tasks.

watchpost