The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of local command execution to perform its primary functions.
SKILL.md contains multiple bash snippets (Sections 1, 2, and 5) that execute git, npm, npx, and gh (GitHub CLI) commands.
The scripts/ci-retry-monitor.py script uses the subprocess module to execute gh api calls for monitoring CI status and triggering reruns.
A Python snippet in SKILL.md (Section 2d) executes git diff via subprocess.check_output to analyze code changes for sensitive data.
[PROMPT_INJECTION]: An indirect prompt injection surface exists in the workflow where untrusted data is processed by the agent or a secondary reviewer agent.
Ingestion points: Untrusted data from the repository (via git diff) is ingested by the primary agent in Section 2d and is intended to be read by a secondary "Independent Closeout Review" agent in Section 3.
Boundary markers: The instructions lack explicit boundary markers or "ignore embedded instructions" warnings to prevent a malicious diff from influencing the agent's behavior during analysis or review.
Capability inventory: The skill environment provides access to Git, npm, and GitHub CLI tools, and the instructions empower agents to make decisions based on the review output.
Sanitization: No sanitization or escaping of the raw diff content is performed before it is presented to the agent for inspection, allowing for potential adversarial instructions hidden in code comments or metadata to be processed.

github-pre-push-gates