The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: SKILL.md, references/command-reference.md, references/security-and-consent.md, and scripts/fabric_check.py contain or recommend the command 'curl -fsSL https://fabric.so/cli/install.sh | sh'. Piping unverified remote scripts directly to a shell is a high-risk pattern for arbitrary code execution.
[COMMAND_EXECUTION]: The bundled scripts scripts/fabric_check.py, scripts/fabric_help_cache.py, and scripts/validate_skill.py utilize the 'subprocess' module to execute external CLI commands and other Python scripts within the package environment.
[DYNAMIC_EXECUTION]: The scripts/validate_skill.py script dynamically identifies and executes other .py scripts located in the scripts/ directory by calling them with 'sys.executable' during its validation routine.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data through 'fabric search' outputs and by reading local files as part of its note-saving and search-summarization workflows.
Boundary markers: Instructions in SKILL.md and references/security-and-consent.md advise the agent to seek user confirmation for destructive actions and broad changes.
Capability inventory: The skill includes capabilities for executing system commands via subprocesses and writing to the local file system.
Sanitization: Mitigation measures are present, including shell argument quoting using 'shlex.quote' in scripts/fabric_command_plan.py and best-effort redaction of API keys and secrets in the output of several diagnostic and note-generation scripts.

fabric-cli