The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data sourced from repository diffs to generate release runbooks and configuration lists. Malicious actors with commit access could potentially craft code comments or string literals designed to influence the agent's output or summary.
Ingestion points: The agent reads from diff_temp.txt, which contains the output of git diff on .java and .sql files.
Boundary markers: The skill instructions in SKILL.md include "Noise Exclusion Rules" to ignore comments, JavaDoc, and log strings, which acts as a partial mitigation.
Capability inventory: The skill utilizes git and grep for data retrieval and temporary file creation.
Sanitization: The skill uses structural pattern matching (e.g., searching for @Value or @XxlJob) to isolate specific data points rather than processing the entire file as free-form text.
[COMMAND_EXECUTION]: The skill requires the execution of shell commands, specifically git diff, to generate the input data. It also instructs the agent to use shell redirection (>) or PowerShell's Out-File to manage temporary files. While these are standard development workflows, they involve direct interaction with the system shell.

sdlc-release-check