aadityakumarsah/hermes/gh-issues

gh-issues

Use for issue-to-PR automation. Prefer gh CLI; fall back to gh api only when a high-level command lacks the needed field.

Arguments

• positional owner/repo: optional; else infer from git remote get-url origin.
• --label <label>: filter.
• --limit <n>: default 10.
• --milestone <title>: filter.
• --assignee <login|@me>: filter.
• --state open|closed|all: default open.
• --fork <owner/repo>: push branches to fork, PR to source.
• --watch: poll issues + reviews.
• --interval <minutes>: default 5.
• --dry-run: list only.
• --yes: no confirmation.
• --reviews-only: skip issue fixing; handle PR reviews.
• --cron: spawn and exit; implies --yes.
• --model <id>: pass to workers when supported.
• --notify-channel <id>: optional final notification target.

Phase 1: resolve repo

git remote get-url origin
if [ -z "${GH_TOKEN:-}" ]; then
  CONFIG_PATH="${OPENCLAW_CONFIG_PATH:-${OPENCLAW_STATE_DIR:-$HOME/.openclaw}/openclaw.json}"
  GH_TOKEN=$(jq -r '.skills.entries["gh-issues"].apiKey // empty' "$CONFIG_PATH" 2>/dev/null || true)
  if [ -n "$GH_TOKEN" ]; then export GH_TOKEN; fi
fi
gh auth status
gh repo view OWNER/REPO --json nameWithOwner,defaultBranchRef

If gh auth status fails and GH_TOKEN is missing, stop and ask for GitHub auth/config.

Derived:

• SOURCE_REPO: issue repo.
• PUSH_REPO: fork if set, else source.
• BASE_BRANCH: source default branch unless user says otherwise.
• PUSH_REMOTE: fork in fork mode, else origin.

Stop on dirty worktree unless user confirms that workers should ignore uncommitted changes.

In fork mode, do not mutate remotes before confirmation or during --dry-run.

Verify auth/read access only:

gh auth token >/dev/null || test -n "${GH_TOKEN:-}"
gh repo view "$PUSH_REPO" --json nameWithOwner
git ls-remote --exit-code origin HEAD

Phase 2: fetch issues

Build filters and fetch:

gh issue list --repo "$SOURCE_REPO" --state open --limit 10 --json number,title,labels,url,body,assignees,milestone

Add --label, --milestone, --assignee, --state, --limit as requested. gh issue list already excludes PRs.

If none found: report no matches. If --dry-run: show compact list and stop.

Phase 3: avoid duplicate work

For each candidate:

gh pr list --repo "$SOURCE_REPO" --search "$SOURCE_REPO#<n>" --state open --json number,url,title,headRefName
gh pr list --repo "$SOURCE_REPO" --head "fix/issue-<n>" --state open --json number,url
gh api "repos/$PUSH_REPO/branches/fix/issue-<n>" >/dev/null

Skip candidates with an open PR, existing branch, or active local claim.

Claim file:

${OPENCLAW_STATE_DIR:-$HOME/.openclaw}/gh-issues-<owner>-<repo>.json

Expire claims older than 2 hours. Create the parent directory before writing.

Phase 4: confirm

Unless --yes or --cron, ask user to choose:

• all
• comma-separated issue numbers
• cancel

After confirmation, in fork mode, configure the push remote before handing work to agents:

gh auth setup-git
git remote get-url fork || git remote add fork "https://github.com/$PUSH_REPO.git"
git remote set-url fork "https://github.com/$PUSH_REPO.git"
git ls-remote --exit-code fork HEAD

Phase 5: spawn workers

Launch up to 8 background workers. Do not block on each worker when --cron.

Before each spawn, write a claim for SOURCE_REPO#<n> with the current ISO timestamp. After a worker reports PR/failure, remove or update the claim. This prevents watch/cron overlap before a branch or PR exists.

Worker prompt must include:

• issue URL, title, body, labels.
• SOURCE_REPO, PUSH_REPO, BASE_BRANCH, PUSH_REMOTE, fork mode.
• target branch fix/issue-<n>.
• required proof and PR body.
• notification route.

Worker instructions:

Use gh and git. Do not handwave.
Checkout/create fix/issue-<n> from BASE_BRANCH.
Implement minimal fix.
Run relevant tests.
Commit with conventional message.
Push to PUSH_REMOTE.
Open PR against SOURCE_REPO BASE_BRANCH.
PR body: What Problem This Solves + Why This Change Was Made + User Impact + Evidence + visible Fixes SOURCE_REPO#<n>.
Report PR URL or failure reason.
Send completion/failure with openclaw message send if route provided.

Use coding-agent launch rules when available.

Phase 6: collect

Poll workers with process or task registry. Report:

• issue number + title.
• status: PR opened, skipped, failed, timed out.
• PR URL or reason.

Notify channel only with final compact summary.

Reviews-only / watch reviews

Discover open PRs:

gh pr list --repo "$SOURCE_REPO" --state open --json number,title,url,headRefName,reviewDecision \
  --jq '[.[] | select(.headRefName | startswith("fix/issue-"))]'

Fetch review threads/comments:

gh pr view <n> --repo "$SOURCE_REPO" --json url,headRefName,comments,reviews
gh api "repos/$SOURCE_REPO/pulls/<n>/comments"
gh api "repos/$SOURCE_REPO/issues/<n>/comments"

Only process fix/issue-* PRs created by this workflow unless the user explicitly named PR numbers. Group actionable comments by PR. Ignore praise, status, duplicates, and already-addressed comments. Spawn one worker per selected/scoped PR, same background rules.

Review worker instructions:

Checkout PR branch.
Read all actionable review comments.
Patch minimal changes.
Run relevant tests.
Commit and push normally; do not force-push unless explicitly told.
Reply to addressed comments with fix + commit/file reference.
Report comments addressed/skipped and proof.

Watch mode

Loop:

1. Fetch issues.
2. Spawn eligible issue workers.
3. Process actionable PR reviews.
4. Sleep --interval.
5. Stop when user says stop.

Keep cumulative summary small.