Find GitHub issues, implement fixes, submit PRs, and track everything.
clawhub install agent-flowforgegh CLI (authenticated): gh auth statusgit configured with your identityclaude CLI (Claude Code): claude --versionnpm install -g @kagura-agent/gogetajob
gogetajob --helpacpx --approve-all claude exec)Never do implementation work in the main session. Always delegate to sub-agents.
CommandWhat it doesgogetajob scan <owner/repo>Discover open issues from a repogogetajob scan --allScan all tracked reposgogetajob feedBrowse available jobsgogetajob check <ref>Deep-inspect an issue before taking itgogetajob start <ref>Take a job — fork/clone/branchgogetajob submit <ref> --tokens NPush + create PR + recordgogetajob followup <ref> --tokens NRecord additional effort on existing workgogetajob syncCheck all PR statuses, flag problemsgogetajob watchSet up automatic sync via crongogetajob statsView overall performance and ROIgogetajob historyView work loggogetajob import <repo>Backfill work_log from GitHub PR history
The full contribution cycle runs as a FlowForge workflow ( workloop). See references/workloop-overview.md for the complete node-by-node breakdown.
Summary:
text
followup → find_work → study → implement → submit → verify → reflect
│ │ │ │
│ └─────────┘ (no good issue? loop back) │
└────────── (has review feedback? handle it first) ──────────┘
To start: flowforge start workloop
Sub-agents delegate all code changes to Claude Code:
bash
cd <repo> && acpx --approve-all claude exec "<task description with full context>"
Task descriptions must include: issue context, reviewer feedback, architecture notes, maintainer preferences from knowledge-base, and a verification suffix:
"Before committing: 1) grep for all test files that import/mock the interfaces you changed, update their mocks. 2) Run the project's test/lint commands. 3) git diff --stat to confirm no files were missed."
Exception: one-line trivial fixes can be done manually.
After each work session, check: did gogetajob, flowforge, or any tool have bugs or friction? If yes:
Before submitting a new PR, check: gh pr list --repo <owner/repo> --author @me --state open
If ≥ 3 open PRs exist, stop. Wait for existing PRs to be reviewed/merged before adding more. Flooding maintainers kills goodwill.
Always pass real token counts from sub-agent session_status:
bash
gogetajob submit <ref> --tokens <actual_count>
gogetajob followup <ref> --tokens <actual_count>
Never estimate. Never guess. No number → don't fill it in.
Before creating any PR:
knowledge-base/projects/<repo>.md (field notes)Run gogetajob sync regularly (or use gogetajob watch for automatic cron).
When sync flags issues:
SignalPriorityActionHuman review commentHighSpawn sub-agent to address, then followupCI failureMediumSpawn sub-agent to fix, then followupBot review (CodeRabbit etc.)LowAddress if substantive, ignore nitpicksPR closed—Read why. If someone else's fix was better, study it
When picking issues from gogetajob feed:
git log --oneline -20 -- <relevant files> to check if already fixed