The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches the Bun installation script from the official well-known service domain (bun.sh).
[REMOTE_CODE_EXECUTION]: Executes the downloaded Bun installation script after verifying its integrity with a SHA256 checksum.
[COMMAND_EXECUTION]: Performs various administrative and setup tasks. • Executes internal binaries located in the ~/.claude/skills/gstack/bin/ directory for configuration management, slug generation, and state tracking. • Automatically updates or creates the CLAUDE.md file in the project root to include skill routing rules, followed by a Git commit. • Utilizes the gh CLI to retrieve Pull Request metadata and the git CLI for branch and diff analysis.
[DATA_EXFILTRATION]: Collects and transmits telemetry data, including skill name, execution duration, and outcome status, to a remote logging endpoint based on user-approved settings.
[PROMPT_INJECTION]: The skill processes untrusted external data, creating an attack surface for indirect prompt injection. • Ingestion points: Reads command output from git diff and gh pr view in the benchmarking workflow. • Boundary markers: No specific delimiters or instruction-isolation markers are present for the ingested data. • Capability inventory: The skill utilizes the Bash, Read, and Write tools to manipulate data and generate report files. • Sanitization: No explicit validation or sanitization is performed on the command output before use in the logic flow.

benchmark