The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from GitLab MR descriptions, comments, and Jira ticket details to generate review findings and verdicts.
Ingestion points: GitLab MR diffs, commits, notes, and Jira ticket summaries, descriptions, and comments (File: SKILL.md).
Boundary markers: The instructions do not specify explicit delimiters for untrusted input within the agent's processing context.
Capability inventory: The skill executes glab CLI commands, writes temporary files to /tmp, and performs POST requests to the GitLab API (File: SKILL.md).
Sanitization: The instructions explicitly mandate keeping arbitrary text out of shell interpolation and using structured arguments for API calls to prevent command injection (File: SKILL.md).
[COMMAND_EXECUTION]: The skill uses the glab CLI to interact with GitLab for capturing snapshots and publishing feedback. While necessary for its primary function, this provides a mechanism for system command execution influenced by MR data (File: SKILL.md).
[DATA_EXFILTRATION]: The skill reads potentially sensitive source code from the repository and internal business requirements from Jira tickets. This data is incorporated into the agent's context and output reports, which is inherent to the tool's function but involves handling sensitive data (File: SKILL.md).

glab-mr-review