The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions include executing multiple shell commands for configuration and repository management, such as gstack-config, gstack-team-init, and git (add, rm, commit) to automate the migration and setup of the gstack environment.
[REMOTE_CODE_EXECUTION]: The skill utilizes eval "$(~/.claude/skills/gstack/bin/gstack-slug 2>/dev/null)" to dynamically execute the output of a local script. This pattern of shell evaluation can be risky if the binary or its output is compromised.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because its primary function is to scrape content from external, untrusted web pages (Step 4 — Prototype phase) and process it as data without explicit sanitization or boundary markers.
Ingestion points: Web content retrieved via $B goto and $B html in the prototype phase of SKILL.md.
Boundary markers: None defined to separate untrusted web data from agent instructions.
Capability inventory: The skill has access to the Bash tool, file system write permissions, and the ability to perform git repository modifications.
Sanitization: No sanitization or validation of the scraped content is performed before the agent processes it to generate JSON output.

scrape