| name | kumo-pr |
| description | Create a branch, fix a kumo-coding-agent skill or context doc, and open a pull request on kumo-ai/kumo-coding-agent. Use when someone says "fix the agent docs", "update a skill", "contribute a fix", "open a PR for kumo-coding-agent", or "improve the agent". |
| argument-hint | [description of what to fix] |
| allowed-tools | ["Bash","Read","Write","Edit","Glob","Grep"] |
Fix kumo-coding-agent and Open a PR
Create a branch, apply fixes to kumo-coding-agent skills or context docs, and
open a pull request for review.
This command works in both Claude Code (interactive) and Codex (headless).
In Codex, provide a description of the fix as the argument โ interactive
prompting is not available.
User input: $ARGUMENTS
Instructions
Step 1: Validate GitHub CLI Auth
Run:
gh auth status
If gh is not authenticated, tell the user to run gh auth login -h github.com
first and stop before creating a branch or making changes.
Step 2: Understand the Fix
If $ARGUMENTS is provided, parse it to understand what needs to change.
If $ARGUMENTS is empty, ask the user (skip in headless mode):
- Which file needs fixing? (skill, context doc, or vertical)
- What's wrong with it?
- Do you know the correct information?
Step 3: Detect Environment and Create a Branch
Generate <slug> from the fix description: lowercase, hyphens, max 50 chars.
Examples: fix-pql-time-units, add-databricks-connector-docs, update-fraud-vertical.
Check whether we're already inside the kumo-coding-agent repo:
git remote get-url origin 2>/dev/null | grep -q "kumo-coding-agent"
If YES (already in the right repo):
git checkout main
git pull --ff-only origin main 2>/dev/null || true
git checkout -b kumo-coding-agent/<slug>
If NO (e.g., running from a project that installed the skills via npx skills add):
WORK_DIR=$(mktemp -d)
git clone --filter=blob:none --depth=1 \
git@github.com:kumo-ai/kumo-coding-agent.git "$WORK_DIR"
cd "$WORK_DIR"
git checkout -b kumo-coding-agent/<slug>
All subsequent steps operate inside $WORK_DIR. Remember it for cleanup.
Step 4: Make the Changes
Apply the requested fixes.
Guidelines:
- Follow existing document structure and conventions
- If updating a context doc, update the Source header date
- If adding new content, keep the same style as surrounding content
- If fixing a claim, check the authoritative source (see
meta/skills/verify-content.md for source locations)
Step 5: Run Verification Checks
After making changes, verify nothing is broken:
python3 -c "import yaml; yaml.safe_load(open('context/_sources.yaml'))" 2>&1
python3 -c "import yaml; yaml.safe_load(open('context/_gaps.yaml'))" 2>&1
grep -oE '`[^`]+\.md`' CLAUDE.md | tr -d '`' | while read f; do
test -f "$f" || echo "MISSING: $f"
done
If any check fails, fix the issue before proceeding.
Step 6: Commit
Stage the changes:
git add .
git status
Review the staged changes. Commit with a descriptive message:
git commit -m "<type>: <description>
<details of what changed and why>"
Where <type> is one of: fix, docs, feat, refactor.
Step 7: Push and Create PR
git push -u origin kumo-coding-agent/<slug>
Create the PR โ always pass --head and --base explicitly so this works
regardless of which directory the command runs from:
gh pr create \
--repo kumo-ai/kumo-coding-agent \
--head kumo-coding-agent/<slug> \
--base main \
--title "<concise title>" \
--body "$(cat <<'EOF'
## Summary
- <1-3 bullet points of what changed>
## Files Changed
- `...`
## Verification
- [ ] YAML files are valid
- [ ] CLAUDE.md routing table references exist
- [ ] Content verified against source (if applicable)
---
*Created via `/kumo-pr` by @AUTHOR*
EOF
)"
Before creating the PR, capture the author's GitHub identity and replace
@AUTHOR in the body:
gh_user=$(gh api user --jq '.login')
If gh is not authenticated, tell the user to run gh auth login -h github.com first.
If a clone tmpdir was created in Step 3, clean it up after the PR is open:
rm -rf "$WORK_DIR"
Step 8: Report
Print:
- The PR URL
- Summary of changes made
- Tell the user: "Your PR has been submitted. A Kumo team member will review it and follow up with you."