The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Setup instructions include installing the Bun runtime via 'curl -fsSL https://bun.sh/install | bash'. While Bun is a well-known development tool, executing remote scripts directly in the shell is a high-risk pattern.\n- [DATA_EXFILTRATION]: The skill implements an opt-in telemetry system that transmits usage metrics (skill names, durations, success/error status) and system metadata (OS, architecture) to a remote Supabase endpoint. When the community tier is active, it generates an installation ID from hashed system 'hostname' and 'whoami' values.\n- [DATA_EXFILTRATION]: The 'bin/dev-setup' script copies '.env' files from the main worktree to the repository root. Since these files frequently contain sensitive secrets or API keys, moving them increases the surface for credential exposure.\n- [COMMAND_EXECUTION]: The preamble and epilogue logic in 'SKILL.md' automatically executes several internal shell scripts ('gstack-update-check', 'gstack-config', 'gstack-telemetry-log') during every invocation to manage session state and check for updates.\n- [EXTERNAL_DOWNLOADS]: The skill performs automated version checks by fetching a version file from a public GitHub repository ('garrytan/gstack'), which is a well-known technology service.\n- [PROMPT_INJECTION]: The 'gstack browse' functionality allows the agent to navigate to and extract content from arbitrary external URLs, creating a surface for Indirect Prompt Injection. \n
Ingestion points: External web content processed via 'text', 'html', and 'accessibility' commands in 'SKILL.md'.\n
Boundary markers: None present; fetched content is treated as direct input for assertions and data extraction.\n
Capability inventory: Subprocess execution of internal scripts in 'bin/', file system writes to '~/.gstack', and network requests via 'curl' for telemetry and updates.\n
Sanitization: No explicit sanitization or filtering of fetched HTML/text content is performed before processing.

gstack