The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The Workflows/Setup.md file contains instructions for the agent to use sudo for installing system packages via apt-get, pacman, and dnf, as well as for copying the downloaded ShellCheck binary into protected system directories like /usr/local/bin/.
[REMOTE_CODE_EXECUTION]: The Setup workflow automates the downloading, extraction, and installation of a binary executable from a remote repository. This binary is then executed to perform the skill's primary linting functions.
[EXTERNAL_DOWNLOADS]: The skill downloads the ShellCheck binary from GitHub Releases and references several third-party GitHub repositories for pre-commit hooks and CI/CD actions. These downloads target a well-known service.
[COMMAND_EXECUTION]: The skill makes extensive use of shell utilities including shellcheck, find, grep, jq, patch, and git. Furthermore, the skill body in SKILL.md specifies an automated curl command to localhost:8888 for sending notifications on every workflow execution.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and analyze shell scripts provided by the user. These scripts are untrusted data and could contain adversarial instructions embedded in comments or strings. If these findings are echoed back into the agent's context during the analysis summary, the agent may inadvertently follow instructions contained within the analyzed content.
Ingestion points: Scripts read and parsed in Workflows/Analyze.md.
Boundary markers: Absent; analysis findings are presented in Markdown blocks without explicit instructions to the agent to ignore any embedded directives.
Capability inventory: File system read/write access (via patch) and network request capabilities (via curl).
Sanitization: No sanitization of the user-provided script content is performed before display or processing.

shellcheck