The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from Jira task titles/descriptions and GitLab MR comments.
Ingestion points: Jira titles (SKILL.md, Mode C/D logic) and GitLab MR comments (SKILL.md, "review the MR !" section).
Boundary markers: The skill requires user confirmation for branch name candidates generated from raw Jira titles and for commits after fixing identified issues.
Capability inventory: git (branching, committing, pushing), glab (MR creation, commenting, merging).
Sanitization: No explicit automated sanitization of Jira titles is mentioned, though the agent is instructed to parse and trim the content before proposing names to the user.
[COMMAND_EXECUTION]: The skill uses git and glab CLI tools extensively to manage the development lifecycle.
Evidence: Commands like git checkout, git push, glab mr create, and glab mr merge are used throughout the file.
Context: These are standard and necessary tools for the skill's stated purpose of automating a GitLab workflow.
[SAFE]: The skill includes several security best practices that mitigate common AI agent risks.
Secret Prevention: It explicitly forbids committing sensitive files such as .env, keys, tokens, or passwords.
Attribution Control: It prohibits AI attribution in commits and Merge Requests to maintain repository integrity and adhere to professional coding standards.
Host Hygiene: It instructs the agent not to write files to global user directories or system locations without explicit user consent, preventing unauthorized persistence or configuration changes.

gitlab-flow