What problem does it solve? Taking a Jira ticket from report to merge request involves many risky steps — reading untrusted issue text, editing code, running tests, committing, pushing, and opening an MR — where a single mistake or injected instruction can cause damage. This Skill runs that entire flow as a gated orchestration where a named human must approve every irreversible step. ## Core Features & Use Cases - Four human approval gates: The workflow stops for named human approval before editing files, before committing, before pushing/opening the MR, and before commenting on Jira. - Prompt-injection defense: Jira issue text is treated as data, never instructions; a deterministic extraction script quarantines imperative text so it never reaches shell commands, branch names, or MR bodies. - Delegated execution: Localization is handed to the progressive-bug-hunter skill and MR publishing to publishing-git-review-requests, while an external policy scaffold enforces allow/confirm/deny rules on every command. - Use Case: A developer says "fix Jira issue DGL-1234 and open an MR" — the agent analyzes the issue, localizes the bug, proposes a plan, waits for approval, implements a capped diff with tests, and opens a properly titled GitLab MR only after each gate is approved. ## Quick Start Ask the agent to run the Jira fix workflow for issue DGL-1234 and open a GitLab merge request with human approval at each gate.
