walk-the-issues
This skill should be used when the user asks to "walk the issues", "work through github issues", "groom and implement issues", "tackle the issue backlog", "start the issue loop", "go through tickets", "work all open issues", "clear the backlog", "process all tickets", or "close out open issues". Orchestrates an end-to-end workflow: groom all open GitHub issues with subagents, then loop through them one-by-one - branching, researching, implementing with swarms, testing, committing, merging, and creating PRs until all issues are complete.
dennisonbertram · 已有静态检查记录
安装与使用
运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。
npx skills add dennisonbertram/developer-skill-pack --skill walk-the-issues技能说明
name: walk-the-issues description: 'This skill should be used when the user asks to "walk the issues", "work through github issues", "groom and implement issues", "tackle the issue backlog", "start the issue loop", "go through tickets", "work all open issues", "clear the backlog", "process all tickets", or "close out open issues". Orchestrates an end-to-end workflow: groom all open GitHub issues with subagents, then loop through them one-by-one - branching, researching, implementing with swarms, testing, committing, merging, and creating PRs until all issues are complete.' version: 0.1.0 user_invocable: true
Walk the Issues
Orchestrate a full end-to-end GitHub issue grooming and implementation workflow. This skill drives two sequential phases: Grooming (analyze all open issues for clarity and completeness), then Implementation Loop (branch → research → swarm implement → test → PR → repeat).
Phase 1: Grooming
Before touching any code, groom every open issue so the implementation loop only encounters well-specified work. Closed issues are skipped entirely — do not re-open or re-analyze them.
Grooming Setup
# Ensure gh CLI is authenticated for your target account
# gh auth switch --user YOUR_GITHUB_USERNAME # Uncomment if needed
gh issue list --state open --json number,title,labels,body,assignees --limit 200
Spawn Grooming Subagents
For each open issue, spawn a read-only Explore subagent (no worktree needed) to analyze it. Run subagents in parallel batches of 4-6.
Each grooming subagent evaluates:
- Already addressed? — Search the codebase for evidence the issue has already been fixed or implemented. Check recent commits, PR history, and relevant source files. If the work is already done, mark the issue
already-resolvedand it will be closed (see Apply Grooming Results below). - Clarity — Is the problem statement unambiguous?
- Acceptance criteria — Are "done" conditions explicit?
- Scope — Is the issue atomic (one concern only)?
- Blockers — Does it depend on another issue?
- Labels — Does it have appropriate labels (bug, enhancement, chore)?
- Effort signal — Rough size (small/medium/large)?
Each subagent outputs its findings to docs/investigations/issue-{number}-grooming.md.
Apply Grooming Results
After subagents complete, the top-level agent applies grooming changes:
- Close already-resolved issues with a comment explaining what existing code/commit addresses it:
gh issue comment {number} --body "Closing — this has already been addressed in {commit/file/PR}. ..." gh issue close {number} - Add labels:
well-specified,needs-clarification,blocked,small,medium,large - Post a comment on unclear issues asking for clarification
- Close duplicates with a reference to the canonical issue
- Update issue body with structured acceptance criteria if missing
# Label example
gh issue edit {number} --add-label "well-specified,small"
# Comment example
gh issue comment {number} --body "Needs clarification: ..."
Only issues labeled well-specified (and still open) enter the implementation loop.
Phase 2: Implementation Loop
Repeat until all well-specified open issues have associated PRs.
Step 1 — Select Next Issue
gh issue list --state open --label "well-specified" --json number,title,labels --limit 1
Pick the lowest-numbered unhandled issue. If none remain, the loop is complete — report a summary and stop.
Step 2 — Create a Working Branch
ISSUE_NUM={number}
SLUG=$(echo "{title}" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g' | cut -c1-40)
BRANCH="issue-${ISSUE_NUM}-${SLUG}"
git checkout main && git pull upstream main
git checkout -b "$BRANCH"
All work for this issue happens on $BRANCH. Never commit directly to main.
Step 3 — Research and Planning
Spawn read-only Explore subagents (no worktree needed) to produce a structured implementation plan. These subagents should:
- Read the issue body, comments, and any referenced files
- Explore the relevant codebase sections
- Identify all files likely to change
- Use Context7 to fetch documentation for any unfamiliar libraries or tools:
Use context7 to look up: {library or tool name} - Identify existing tests covering affected code
- Output plan to
docs/plans/issue-{number}-plan.mdusing the project plan template
Plan must include:
- Summary of changes required
- Files to create/modify
- Testing strategy (unit + regression)
- Risk areas / edge cases
- Commit strategy (what logical commits to make)
- Library Research section (context7 findings, if any)
Step 4 — Regression Tests First
Before implementing, write regression tests for every code path that will change.
Spawn a general-purpose subagent with worktree isolation to:
- Read
docs/plans/issue-{number}-plan.md - Identify existing test files covering affected areas
- Add regression tests that will fail until the implementation is correct
- Commit the failing tests:
git add <specific test files> git commit -m "test(#{number}): add regression tests for {area}"
This ensures the test suite is red before implementation begins.
Step 5 — Implement with Swarms
Use a swarm (2–4 agents in parallel per wave) for implementation. Each agent works in its own worktree (isolation: "worktree").
阅读完整来源 ↗Review Policy — This project is internal software under the user's control, not deployed in an adversarial environment. The default swarm skill requires 3 consecutive passing reviews across Security/UX/Correctness perspectives. That is not required here. A single round of review is sufficient. The hard requirements are:
- All tests pass (
go test ./...+-race)- Regression tests are written for every changed code path
- New functionality has unit tests
- Coverage gate holds (80% minimum