92
GitHub Stars
1
Bundled Files
2 months ago
Catalog Refreshed
3 months ago
First Indexed
Readme & install
Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.
Installation
Preview and clipboard use veilstrat where the catalogue uses aiagentskills.
npx veilstrat add skill jmagly/aiwg --skill issue-driven-ralph- SKILL.md6.3 KB
Overview
This skill orchestrates issue-driven Ralph loops that actively work issues and post structured status updates to issue threads. It turns the issue tracker into a two-way collaboration surface, scanning for human feedback each cycle and adapting work accordingly. The agent respects safety limits, runs tests before declaring completion, and supports interactive or guided runs.
How this skill works
When triggered by natural language (e.g., "tackle issue 17" or "work on the bug backlog"), the skill extracts issue numbers, filters, and flags like --interactive or --guidance and invokes the /address-issues command with those parameters. Each ralph cycle performs work, posts a structured status comment to the issue thread, and scans new thread comments to classify and incorporate human feedback into the next cycle. The loop repeats until completion criteria or max cycles are reached, with optional batching or parallel strategies when requested.
When to use it
- Automatically progress open issues and keep stakeholders informed
- Work through a bug backlog or specific issue numbers
- Run interactive sessions for human approval between cycles
- Prioritize or guide the loop with upfront instructions (e.g., security first)
- Batch related issues or run parallel work when context budget allows
Best practices
- Provide clear guidance when you want a specific focus (use --guidance text)
- Use --interactive for manual checkpoints and go/no-go decisions
- Keep max cycles reasonable (default 6) and increase only when necessary
- Always review cycle status comments and respond in-thread to steer work
- Avoid destructive git operations; the agent never force-pushes
Example use cases
- "Tackle issue 17" → run /address-issues 17 and post cycle updates to issue 17
- "Work on the bug backlog" → run /address-issues --filter "status:open label:bug"
- "Address issues 17, 18, 19 interactively" → pause between issues for human approval
- "Go through all open issues, give each one up to 8 cycles" → /address-issues --all-open --max-cycles 8
- "Fix the open bugs, focus on security issues first" → pass --guidance "Security issues are top priority"
FAQ
Each cycle the agent scans thread comments, classifies them (feedback, question, approval, correction), and incorporates relevant items into the next cycle; it never ignores human input.
What marks an issue as resolved?
An issue is resolved when implementation is complete, tests pass, documentation is updated if needed, all thread feedback is addressed, and there are no unresolved blocker comments.