End-to-end issue pipeline: plan → implement → review → merge. Dispatches parallel sub-agents with worktree isolation. Usage: /issue-pipeline 566 567 568 [--skip-plan] [--no-merge]
Scanned 9/2/2026
Install to Claude Code
npx -y skills add fagemx/edda --skill issue-pipeline --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Issue Pipeline?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/fagemx-issue-pipeline)More formats (shields.io, HTML) on the badges page.
---
name: issue-pipeline
description: "End-to-end issue pipeline: plan → implement → review → merge. Dispatches parallel sub-agents with worktree isolation. Usage: /issue-pipeline 566 567 568 [--skip-plan] [--no-merge]"
---
# Issue Pipeline Skill
You orchestrate the full lifecycle of GitHub issues through parallel sub-agents. Each phase runs all issues concurrently in isolated worktrees.
## Usage
```
/issue-pipeline <issue-numbers...> [flags]
```
**Flags:**
- `--skip-plan` — Skip plan phase, go straight to implement (issues already have plans)
- `--no-merge` — Stop after review, don't auto-merge
There is deliberately no `--skip-review` flag: review can never be skipped. Merging
requires a final current-head LGTM (see Phase 4); work that fails or lacks review
routes through the `pr-review-loop` skill instead.
## Prerequisites
This skill depends on these sub-skills being installed in `.claude/skills/`:
- `issue-plan` — Deep-dive planning (research → innovate → plan)
- `issue-action` — Implementation from plan to PR
- `pr-review-loop` — Iterative review with auto-fix
If any are missing, inform the user and suggest installing them from the karvi starter-kit.
## Pipeline Phases
### Phase 1: Plan (parallel)
For each issue, launch a sub-agent with `isolation: "worktree"` and `run_in_background: true`:
```
prompt: |
You are working on GitHub issue #{number} for this project at {project_root}.
Load and execute the issue-plan skill by reading {project_root}/.claude/skills/issue-plan/SKILL.md
and following its instructions for issue #{number}.
IMPORTANT: You are in a worktree. Do NOT modify source code — planning only.
```
**Wait for all agents to complete.** Report results to user as a table.
### Phase 2: Implement (parallel)
For each issue, launch a sub-agent with `isolation: "worktree"` and `run_in_background: true`:
```
prompt: |
You are working on GitHub issue #{number} for this project at {project_root}.
Load and execute the issue-action skill by reading {project_root}/.claude/skills/issue-action/SKILL.md
and following its instructions for issue #{number}.
The plan has been posted as a comment on the issue — read it with `gh issue view {number} --comments`.
IMPORTANT: You are in a worktree. Pull latest main first: `git pull origin main`.
```
**Wait for all agents to complete.** Collect PR URLs. Report results to user.
### Phase 3: Review (parallel)
For each PR created in Phase 2, launch a sub-agent with `isolation: "worktree"` and `run_in_background: true`:
```
prompt: |
You are reviewing PR #{pr_number} for this project at {project_root}.
Load and execute the pr-review-loop skill by reading {project_root}/.claude/skills/pr-review-loop/SKILL.md
and following its instructions for PR #{pr_number}.
IMPORTANT: You are in a worktree. If fixes are needed, make them on the PR's branch, commit, and push.
```
**Wait for all agents to complete.** Report verdicts.
### Phase 4: Merge
**Merge preconditions** (`pr.merge-policy`): a PR may be merged only when all of the
following hold —
- A final **current-head** LGTM from review (Phase 3, latest pushed SHA)
- P0 = 0 and P1 = 0 blocking findings
- All required checks green
For each PR meeting every precondition:
```bash
gh pr merge {pr_number} --squash
```
Never pass `--delete-branch` — branches and worktrees are never deleted (see
`.claude/CLAUDE.md`). If a PR does not meet the preconditions, do not merge it;
route it back through the `pr-review-loop` skill and report that it is blocked.
Report final status table.
## Execution Rules
1. **All sub-agents use `isolation: "worktree"`** — never work on main directly
2. **All sub-agents use `run_in_background: true`** — maximize parallelism
3. **Launch all agents for a phase in a SINGLE message** (one tool call per agent, all in parallel)
4. **Wait for ALL agents in a phase before starting the next phase**
5. **Report a status table after each phase** — issue number, status, key details
6. **If any agent fails, report it but continue with the rest** — don't block the pipeline
7. **Never delete branches or worktrees** — worktrees, branches, and sources are
never deleted (see `.claude/CLAUDE.md`); `git worktree remove` and
`git worktree prune` are forbidden; leave cleanup to the operator
## Status Table Format
After each phase:
```
| Issue | Phase | Status | Details |
|-------|-------|--------|---------|
| #566 | plan | ✅ | Plan posted to issue |
| #567 | plan | ✅ | Plan posted to issue |
| #568 | plan | ❌ | Agent error: ... |
```
## Error Handling
- **Agent fails**: Mark as ❌, continue with other issues. Report at end.
- **PR creation fails**: Try to identify branch, suggest manual fix.
- **Review finds issues**: Review loop handles fix + re-review automatically.
- **Merge conflict**: Report conflict, skip merge for that PR, suggest manual resolution.
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!