The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill facilitates the download and execution of a shell script from an external URL which could be manipulated.
Evidence: curl -s -S -L https://gam-shortn.appspot.com/gam-install -o /tmp/gam-install.sh && bash /tmp/gam-install.sh in SKILL.md.
[PERSISTENCE]: The skill uses crontab to schedule recurring tasks on the system.
Evidence: echo "0 */4 * * * curl ... " | crontab - in references/details.md. This command replaces the existing user crontab, potentially leading to data loss of other scheduled tasks while establishing a persistent execution mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data which is then used as arguments in shell commands and API calls.
Ingestion points: users.csv in SKILL.md and /tmp/okta-user-ids.txt in references/details.md.
Boundary markers: Absent; delimiters or instructions to ignore embedded commands are not used when processing the files.
Capability inventory: The skill has extensive capabilities including executing cloud management tools (gam, az, aws, gh), performing network requests, and modifying system configuration.
Sanitization: No sanitization, escaping, or validation of the input data is implemented before use in commands.
[COMMAND_EXECUTION]: The skill relies heavily on shell command execution and interaction with various CLI tools to perform its functions.
Evidence: Extensive use of gam, curl, az, gh, and aws CLI tools across all files.
[METADATA_POISONING]: The skill contains inconsistent author metadata.
Evidence: The YAML frontmatter in SKILL.md identifies the author as devops-skills, which is inconsistent with the provided skill author sickn33.
[DATA_EXFILTRATION]: The skill writes potentially sensitive identity and access data to shared temporary directories.
Evidence: Exporting user lists and group memberships to /tmp/access-review-groups.csv, /tmp/okta-users.csv, and /tmp/okta-access-review.csv in references/details.md.

identity-access-management