The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/audit_consumer.py executes the git command via subprocess.run to verify whether project-specific files are correctly tracked in the version control system.
Evidence: The script calls subprocess.run(['git', '-C', str(root), 'ls-files', '--error-unmatch', tracked_input], ...).
Assessment: This is a standard operation for developer tooling; the use of an argument list without a shell (shell=True is absent) mitigates command injection risks.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection because it ingests and processes untrusted files from the repository being audited.
Ingestion points: The audit script reads Cargo.toml, Cargo.lock, Dioxus.toml, justfile, and source files (main.rs, lib.rs) from the user's repository.
Boundary markers: None explicitly implemented in the reporting phase, though the script performs logical checks rather than raw interpolation.
Capability inventory: The script can read local files and execute git. It does not perform network operations or write to the file system.
Sanitization: The script uses robust standard library parsers (tomllib for TOML and xml.etree.ElementTree for XML sitemaps) and specific string matching for code validation, reducing the likelihood of processing malicious payloads as instructions.

use-stayhydated-github-pages