Run the fix, ship, and roast cycle for an open pull request until the latest roast reports zero findings. Use when the user asks to repeat review fixes until clean or run the review loop.
Installs into .claude/skills of the current project.
Are you the author of Warpath?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/karloows-warpath)
---
name: warpath
description: Run the fix, ship, and roast cycle for an open pull request until the latest roast reports zero findings. Use when the user asks to repeat review fixes until clean or run the review loop.
---
# Warpath Workflow
Use this skill when the user wants Orchraft to keep cycling through review
findings instead of stopping after one ship or roast pass.
## At A Glance
1. Read the live branch and pull-request state with `runes`.
2. If findings exist, hand them to `Forgehand` for fixes.
3. When fixes are ready, hand off to `Raid Captain` / `ship`.
4. Run `roast` against the current pull-request head, including the initial
head when no prior roast exists.
5. If findings remain, repeat from step 2; otherwise the latest roast is clean.
6. Recommend `Haulmaster` / `land`; do not merge as part of this loop.
## Completion
The loop is complete only when the latest roast pass ran against the current
PR head and reported zero findings. A successful ship, passing checks, or zero
unresolved threads from an older head is not completion.
If a finding is declined, keep it visible and stop for the user's decision;
declining is not zero findings. If checks fail, the PR disappears, or the
review cannot be posted, stop and report the blocking state.
## Approval
Read `context/policies/approval-policy.md` before starting. This workflow is
an orchestrator, not a permission grant. If the target repo's verified
Autonomous Mode names the action under `autonomous.warpath`, continue through
that action without asking again during this user-started run. Otherwise pause
before every commit, push, PR update, review, thread resolution, or other
mutation. Landing always remains a separate, explicit user request.
Do not create a stored loop state file. Re-read live git and GitHub state after
each handoff so a resumed run cannot mistake a stale roast for a clean one.
## Handoff
Unless the user asked for plain output, open with a short orc-voice line,
speaking as the Warpath from `context/personality.md` (the target repo's
copy when it exists, otherwise `${CLAUDE_PLUGIN_ROOT}/context/personality.md`),
and state the current cycle.
Report every cycle's finding count and PR head. When clean, hand off to
`Haulmaster` / `land` instead of invoking it.
