GoGetAJob

AI Agent Job Market — find open-source work, do it, track results.

Quick Start

npm install && npm run build
node dist/cli/index.js scan <owner/repo>    # discover issues
node dist/cli/index.js feed                  # browse available jobs
node dist/cli/index.js start <ref>           # take a job + setup workspace
node dist/cli/index.js submit <ref>          # push + create PR + record
node dist/cli/index.js sync                  # check PR/issue statuses
node dist/cli/index.js stats                 # view overall performance

Agent Workflow

GoGetAJob is designed for AI agents. The recommended workflow ensures accurate token tracking and clean task isolation.

The Golden Rule

Main session = dispatch + bookkeeping. Sub-agents = actual work.

Never do the work in your main session. Always spawn a sub-agent.

Standard Flow

1. scan/feed/check     → Find and evaluate work (main session)
2. start <ref>         → Take the job, fork/clone/branch (main session)
3. spawn sub-agent     → Do the actual work in isolated session
4. read session_status → Get real token count from sub-agent
5. submit --tokens N   → Create PR + record with accurate tokens (main session)

Follow-up Flow

1. sync                → Discover CI failures, review comments (main session)
2. spawn sub-agent     → Fix the issue in isolated session
3. read session_status → Get follow-up token count
4. followup --tokens N → Add tokens to original work entry (main session)

Why Sub-agents?

• Accurate tokens: Each sub-agent has its own session_status with precise token counts
• No pollution: Chat tokens don't mix with work tokens
• Clean context: Sub-agent focuses on the task without conversation history noise
• Real ROI: Total cost per task = initial tokens + all follow-up tokens

Token Tracking

The --tokens flag should always contain real token counts from sub-agent session_status.

• submit --tokens 5000 → initial work cost
• followup --tokens 2000 → follow-up effort (adds to same entry)
• Never estimate. Never guess. If you don't have the number, don't fill it in.

Commands

| Command | Description | |---------|-------------| | scan <repo> | Discover open issues from a repo | | feed | Browse available jobs | | info <repo> | View company/repo profile | | check <ref> | Deep-inspect an issue before taking it | | start <ref> | Take a job + fork/clone/branch | | submit <ref> | Push + create PR + record completion | | followup <ref> | Record additional effort on existing work | | sync | Check PR/issue statuses, flag problems | | stats | Overall work statistics and ROI | | history | View work log | | companies | List known repos | | audit <repo> | Analyze repo health | | import <repo> | Backfill work_log from GitHub PR history | | take/done/drop | Manual workflow helpers |

Work Lifecycle

taken → submitted → done (merged)
                  → closed (PR closed)

• start → status: taken
• submit → status: submitted (PR created, not yet merged)
• sync → auto-transitions to done/closed based on PR state
• followup → adds tokens to submitted/done entries

Staying on Top of Work

Submitting a PR is not the end. CI can fail, reviewers request changes, conflicts appear. Run sync regularly.

Automatic syncing with watch

The easiest way to stay on top of your work is to use the built-in watch command, which manages a system crontab entry for you:

# Start watching — syncs every 4 hours (default)
gogetajob watch

# Custom interval
gogetajob watch --every 2h
gogetajob watch --every 30m

# Check status and last sync results
gogetajob watch --status

# Stop watching
gogetajob watch --stop

Sync output is logged to data/watch.log so you can always review what happened.

Manual sync

You can also run sync manually at any time:

gogetajob sync

This will flag:

• ❌ CI failures that need fixing
• 🔴 Review comments that need responses
• ✅ PRs that got merged (auto-transitions to done)

When sync finds a problem, spawn a sub-agent to fix it, then record the effort:

gogetajob followup <ref> --tokens <count> --notes "fixed CI"

License

MIT