The Agent Skills Directory

[SAFE]: The skill functions as a read-only dashboard for auditing the state of setup tools within a project. It is designed to verify version stamps and file integrity without modifying any files or environment settings.
[SAFE]: The skill operates under a restricted capability model. Its allowed-tools are limited to Read, Bash, Glob, and Grep, and it explicitly excludes Write, Edit, and Agent tools, ensuring that it cannot perform unauthorized changes to the project.
[SAFE]: The static analysis flag regarding concealment is a false positive. The skill's instructions actually enforce transparency by requiring the agent to provide a mandatory explanation to the user about why it does not auto-execute fixes (to prevent AI context degradation), and instead provides manual commands for user review.
[SAFE]: While the skill accesses configuration files in the user's home directory (~/.claude), this is done solely to audit global tool settings. The skill does not possess network tools (like curl or wget), which prevents the exfiltration of any sensitive data it might read.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection due to its role in processing external project files, but the risk is minimized by its design.
Ingestion points: Project-specific artifacts and metadata files located in the .claude/ directory and global configurations in ~/.claude/.
Boundary markers: The instructions mandate a strict output structure (a PLAN block, counts, and tables) and explicitly tell the agent to translate only prose while keeping commands and paths verbatim.
Capability inventory: The skill uses Read, Bash, Glob, and Grep; it lacks the ability to write files, edit content, or spawn sub-agents.
Sanitization: Data is extracted using specific regex and text-processing tools (grep, sed, awk) to pull metadata fields rather than interpreting the file content as instructions.

setup-status