The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's setup routine downloads the Bun installation script from the official well-known service URL https://bun.sh/install. The script implements a security best practice by verifying the downloaded file against a hardcoded SHA256 checksum (bab8acfb046aac8c72407bdcce903957665d655d7acaa3e11c7c4616beae68dd) before execution.
[COMMAND_EXECUTION]: The preamble and core workflow execute various local helper binaries located in the vendor's directory (~/.claude/skills/gstack/bin/). These tools manage configuration, telemetry, project slugs, and timeline logging as part of the skill's operational framework.
[DATA_EXFILTRATION]: The skill implements a telemetry system that collects metadata such as skill duration, outcomes, and repository names. This system is transparently managed: users are prompted via AskUserQuestion to choose their telemetry level (Community, Anonymous, or Off), and the data is stored locally in ~/.gstack/analytics/ before any remote transmission.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection because it ingests untrusted data from external websites (via links, text, and console commands) during the monitoring phases.
Ingestion points: External site links and page text are read using the browse daemon in Phases 3, 4, and 5.
Boundary markers: The instructions do not specify explicit delimiters for the data ingested from external URLs.
Capability inventory: The skill has access to Bash, Read, and Write tools to perform monitoring and reporting.
Sanitization: No explicit sanitization or filtering of the external site content is described in the workflow.

canary