The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read, fix, and format justfile content provided by the user. Maliciously crafted recipes or comments within these files could attempt to influence the agent's behavior during the analysis or modification phase.
Ingestion points: The script scripts/check_justfiles.py discovers and reads justfile and *.just files from the project directory.
Boundary markers: The skill instructions mandate the use of just --fmt --check and just --dry-run to validate syntax and verify command text without full execution, providing a layer of oversight.
Capability inventory: The agent has the ability to execute shell commands via the just tool and perform file writes to the local filesystem.
Sanitization: The skill relies on the native just parser for syntax validation and instructs the agent to manually review dry-run outputs and seek user confirmation for sensitive tasks.
[COMMAND_EXECUTION]: The helper script scripts/check_justfiles.py executes the local just binary to perform formatting checks.
Evidence: The validate function in scripts/check_justfiles.py uses subprocess.run with a list of arguments ([executable, "--fmt", "--check", "--justfile", str(path)]) and shell=False (default), which is a secure method for executing external processes.
[SAFE_PRACTICES]: The skill explicitly advises against security pitfalls common in command runners.
It instructs the agent to avoid bare interpolation {{ arg }} to prevent shell injection, recommending "$@" or quote() instead.
It warns against using {{ env('FOO') }} for secrets to prevent them from being leaked in dry-run outputs or logs, recommending environment variable access ($FOO) instead.

write-justfile