The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a local Bash script (scripts/storage-audit.sh) to perform disk usage analysis. This script utilizes standard system utilities such as du, find, git, and lsof. All operations are scoped to the user-provided root directory, and the script uses absolute paths to prevent directory traversal or command injection via relative paths.
[INDIRECT_PROMPT_INJECTION]: The skill processes filesystem metadata, including file names and directory structures, which are technically untrusted inputs. However, the implementation mitigates risk by validating that paths contain no control characters (via has_unsupported_name), using structured JSON for data exchange, and requiring the agent to wait for explicit user selection of deterministic IDs before recommending any actions.
[SAFE]: The skill adheres to the principle of least privilege by maintaining a strictly read-only posture regarding the audited filesystem. It explicitly forbids the autonomous execution of destructive commands like rm or mv, instead providing human-readable commands in stdout for manual review and execution by the user. The safety contract explicitly prohibits network access and package manager invocations.

storage-audit