Automating GitHub Issues with OpenClaw AI

# Confirm gh is authenticated
gh auth status

# Confirm OpenClaw can use the shell tool
openclaw run "List the last 3 open issues in openclaw-ai/openclaw" --allow-shell

# Fetch all unlabeled open issues and triage them
openclaw run --allow-shell "
Fetch all open issues in openclaw-ai/openclaw that have no labels (use gh issue list).
For each issue, read its body and title, then apply the most appropriate label from:
  bug, enhancement, documentation, question, wontfix, good first issue, help wanted.

Use: gh issue edit <number> --add-label <label> --repo openclaw-ai/openclaw

Print a table: issue number | title | label applied.
"

#!/usr/bin/env bash
# scripts/triage-issues.sh
set -euo pipefail

REPO="${GITHUB_REPO:-openclaw-ai/openclaw}"

echo "=== Triaging unlabeled issues in $REPO ==="
openclaw run --allow-shell "
Fetch all open issues with no labels in $REPO using gh issue list.
For each issue: read its body, classify it as one of:
  bug / enhancement / documentation / question / good first issue
Apply the label with: gh issue edit  --add-label  --repo $REPO
Print: [#number] title → label
" --model gpt-4o-mini

echo "Done."

openclaw run --allow-shell "
Fetch open issues in openclaw-ai/openclaw that:
  1. Are labeled 'question' AND have zero comments, OR
  2. Body contains 'how do I' or 'how to'

For each such issue:
  a. Read the full issue body.
  b. Check our documentation at https://openclaw-wiki.top to find a relevant answer.
  c. Post a helpful comment using:
     gh issue comment <number> --body '...' --repo openclaw-ai/openclaw

The comment should:
  - Greet the author by their GitHub handle
  - Answer the question directly with a code snippet or a documentation link
  - End with 'Feel free to reopen if this doesn't solve your problem.'

Print: commented on issue #<number> — <title>
"

cat prompts/issue-response.txt
# You are a helpful OpenClaw maintainer. Always be friendly and technically accurate.
# If you cannot find a definitive answer, say so and ask clarifying questions.
# Never close an issue; only comment. Never apply labels unless explicitly told to.

openclaw run --prompt-file prompts/issue-response.txt --allow-shell \
  "Handle new unanswered questions in openclaw-ai/openclaw" \
  --model gpt-4o

#!/usr/bin/env bash
# scripts/errors-to-issues.sh
# Usage: ./errors-to-issues.sh /var/log/app-error.log

LOG_FILE="${1:-/var/log/app-error.log}"
REPO="${GITHUB_REPO:-openclaw-ai/openclaw}"
CUTOFF="1 hour ago"

# Extract errors from the last hour
RECENT_ERRORS=$(journalctl --since="$CUTOFF" -p err 2>/dev/null \
  || tail -n 200 "$LOG_FILE")

openclaw run --allow-shell "
Analyse these application errors and decide which ones are worth filing as GitHub issues.
Errors: $RECENT_ERRORS

Rules:
- Skip errors that already have an open issue (use: gh issue list --search 'error text')
- Skip errors that are known false positives (rate limits, transient network errors)
- For each new, actionable error:
  a. Write a clear title: '[BUG] <short description>'
  b. Write a body with: summary, stack trace (code-fenced), steps to reproduce if obvious, environment info
  c. Create the issue: gh issue create --title '...' --body '...' --label bug --repo $REPO
  d. Print: created issue #<number> for <error type>
"

#!/usr/bin/env bash
# scripts/daily-digest.sh
set -euo pipefail

REPO="${GITHUB_REPO:-openclaw-ai/openclaw}"
DATE=$(date +%Y-%m-%d)
OUTPUT="digests/issues-$DATE.md"
mkdir -p digests

openclaw run --allow-shell "
Fetch all open issues in $REPO using gh issue list --state open --limit 100.
Generate a team digest in Markdown with these sections:

## 🔴 Critical (labeled: bug, priority:high) — needs action today
## 🟡 Open PRs Ready for Review
## 📋 New Issues This Week
## 📊 Stats: total open, opened this week, closed this week

Use gh commands to fetch additional data as needed.
Format each issue as: #number — title (author, age, labels)
Keep each section short and scannable.
" > "$OUTPUT"

echo "=== Daily Digest saved to $OUTPUT ==="
cat "$OUTPUT"

# After generating the digest, post to Slack
DIGEST=$(cat "$OUTPUT")
curl -s -X POST "$SLACK_WEBHOOK_URL" \
  -H 'Content-type: application/json' \
  --data "{\"text\": \"*Daily GitHub Issues Digest — $DATE*\n\`\`\`${DIGEST:0:2900}\`\`\`\"}"

# .github/workflows/issue-automation.yml
name: Issue Automation

on:
  schedule:
    - cron: '0 9 * * 1-5'   # 09:00 UTC, Mon–Fri
  workflow_dispatch:          # allow manual trigger

jobs:
  triage:
    runs-on: ubuntu-latest
    permissions:
      issues: write
      contents: read
    steps:
      - uses: actions/checkout@v4

      - name: Install OpenClaw
        run: pip install openclaw

      - name: Configure OpenClaw
        run: |
          mkdir -p ~/.openclaw
          echo "provider: openai" > ~/.openclaw/config.yaml
          echo "model: gpt-4o-mini" >> ~/.openclaw/config.yaml
          echo "allow_shell: true" >> ~/.openclaw/config.yaml
        env:
          OPENCLAW_API_KEY: ${{ secrets.OPENCLAW_API_KEY }}

      - name: Triage unlabeled issues
        run: bash scripts/triage-issues.sh
        env:
          GITHUB_REPO: ${{ github.repository }}
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}

      - name: Generate daily digest
        run: bash scripts/daily-digest.sh
        env:
          GITHUB_REPO: ${{ github.repository }}
          GH_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          SLACK_WEBHOOK_URL: ${{ secrets.SLACK_WEBHOOK_URL }}

openclaw run --allow-shell "
Find all open issues in $REPO that:
  - Have no activity (comments or label changes) in the last 90 days
  - Are NOT labeled 'pinned' or 'roadmap'

For each stale issue:
  1. Post: 'This issue has been inactive for 90 days. Closing as stale — reopen if still relevant.'
     Command: gh issue comment <n> --body '...' --repo $REPO
  2. Close it: gh issue close <n> --repo $REPO

Print: closed #<n> — <title>
"

openclaw run --allow-shell "
Fetch the 50 most recent open issues in $REPO using gh issue list.
Group issues by semantic similarity (read their bodies carefully).
For any group of 2+ issues that appear to describe the same problem:
  - On the newest duplicate, post a comment: 'Possible duplicate of #<oldest number>'
  - Apply the 'duplicate' label: gh issue edit <n> --add-label duplicate

Print: possible duplicate pairs found.
" --model gpt-4o