Autopilot mode (issue → PR)
The full three-skill chain — assigner → executor → reviewer — run headlessly in CI as separate jobs, one per skill, each on its own Jira identity. A GitHub issue goes in; an open, reviewed PR comes out. Nothing merges automatically — that stays a human act on both the Jira and GitHub sides.
| Workflow file | Trigger | Branch / PR target | Deep dive |
|---|---|---|---|
demo-claude-feature-flow.yml | Comment /make-feature on an issue | feature/<KEY>-<slug> off <DEFAULT_BASE_BRANCH>; PR into <DEFAULT_BASE_BRANCH> | ci-feature-flow-demo.md |
demo-fcc-nvidia-nim-feature-flow.yml | Comment /fcc-make-feature on an issue | Same as above | Same skills, same job shape as demo-claude-feature-flow.yml — the only difference is the model backend (Free Claude Code + NVIDIA NIM instead of the Claude Code CLI). See ci-feature-flow-demo.md. |
demo-claude-hotfix-flow.yml | Comment /make-hotfix on an issue | hotfix/<KEY>-<slug> off origin/<PRODUCTION_BRANCH>; PR into <PRODUCTION_BRANCH> | ci-hotfix-flow-demo.md |
No job in this category declares an environment, so a run that passes the OWNER comment guard goes start to finish with no approval pause (see GITHUB-AUTOMATIONS.md §3 for the gating convention shared by every workflow that runs a coding assistant).