--- name: mission-go description: "Launch the Ralph execution loop to process mission objectives. Use when ready to execute a planned mission with the Pod\/Worker\/Inspector\/Analyst cycle." --- # \/mission-go - Ralph Execution Loop Launch the Ralph execution loop to process objectives for an active mission. Ralph spawns fresh Pods for each objective, maintaining state in SQLite while avoiding context rot. --- ## Purpose The Ralph loop is the execution engine of Space-Agents. It: 1. Picks up pending objectives from SQLite (ordered by priority) 2. Spawns a fresh Pod for each objective 3. Monitors Pod exit codes to determine success\/failure 4. Creates alerts on blockers or critical failures 5. Continues until mission complete or critical halt ### Core Principle > **"Agents are compute, not memory."** Each Pod is spawned fresh with no memory of previous objectives. State persists in SQLite. This prevents context rot and enables indefinite scaling. --- ## When to Use Run `\/mission-go` when: - A mission has been planned with objectives in SQLite - The mission status is "active" - You're ready to begin automated execution **Prerequisites:** - Space-Agents must be installed (`\/install`) - A mission must exist in the database - The mission must have status "active" - At least one objective must be "pending" ### Activating a Staged Mission When a mission is planned via `\/mission-brief`, it starts with status "staged" and files in: ``` .space-agents\/missions\/staged\/\/ ``` Before launching, you must: 1. Update the mission status to "active" in SQLite 2. Move mission files from `staged\/` to `active\/` 3. Create the handovers folder for inter-Pod context sharing ```sql UPDATE missions SET status = 'active' WHERE id = ''; ``` ```bash mkdir -p .space-agents\/missions\/active\/\/handovers mv .space-agents\/missions\/staged\/\/* .space-agents\/missions\/active\/\/ rmdir .space-agents\/missions\/staged\/ ``` --- ## Mode Selection Before launching Ralph, choose an execution mode: ### Visible Mode (`--visible`) - Recommended Ralph launches inside mprocs, spawning each Pod as a visible panel. Use when: - You want to watch Pods execute in real-time - Debugging or demonstrating the system - You need to see what each Pod is doing **IMPORTANT:** Visible mode requires manual execution. Claude Code cannot run mprocs through the Bash tool - it must run in your actual terminal to display the TUI. When the user selects visible mode, provide the command for them to copy\/paste: ``` HOUSTON: Run this in your terminal to launch visible mode: cd \/path\/to\/project && bash .claude\/plugins\/cache\/space-agents\/space-agents\/\/skills\/mission-go\/scripts\/ralph.sh MSN-XXX --visible (Or if using local skills: bash skills\/mission-go\/scripts\/ralph.sh MSN-XXX --visible) ``` The user runs this command manually and watches the mprocs screen. Each Pod spawns as a new panel with live output. ### Attended Mode (`--attended`) Ralph runs in the foreground with full output. Use when: - Debugging or first-time execution - You want to watch Pod progress in real-time - You need to intervene quickly if issues arise ```bash bash scripts\/ralph.sh --attended ``` ### Background Mode (default) Ralph runs with minimal output. Use when: - Executing a well-tested workflow - Running overnight or unattended - You'll check status via `\/capcom` ```bash bash scripts\/ralph.sh ``` --- ## Instructions When `\/mission-go` is invoked, follow these steps: ### Step 1: Verify Prerequisites Query SQLite to find active missions: ```sql SELECT m.id, m.title, (SELECT COUNT(*) FROM objectives WHERE mission_id = m.id AND status = 'pending') as pending_count FROM missions m WHERE m.status = 'active'; ``` If no active missions: ``` HOUSTON: No active missions found. Use \/mission-brief to create a mission first. ``` ### Step 2: Confirm Mission Selection If multiple active missions exist, present options to the user: ``` Active missions found: [1] MSN-001: JWT Token Management (VOY-001: Auth System) - 5 objectives pending [2] MSN-002: User Registration (VOY-001: Auth System) - 3 objectives pending Which mission should I launch? ``` ### Step 3: Select Mode Ask the user: ``` Launch mode: [1] Attended - Full output, watch in real-time [2] Background - Minimal output, check via \/capcom Select mode: ``` ### Step 4: Launch Ralph Execute the ralph.sh script: ```bash # Attended mode bash scripts\/ralph.sh --attended # Background mode bash scripts\/ralph.sh ``` ### Step 5: Monitor and Report After launch: **Attended mode:** Output streams directly to terminal. **Background mode:** Report: ``` HOUSTON: Ralph loop launched for mission MSN-001. Monitor progress with \/capcom. Notifications will appear on completion or critical failure. ``` --- ## Exit Codes Ralph reports these exit codes: | Code | Meaning | Action | |------|---------|--------| | 0 | Mission complete | All objectives done. Celebrate. | | 1 | Mission failed | Critical alert or all objectives failed. Check alerts. | | 2 | Configuration error | DB missing, mission not found, etc. Run \/launch first. | --- ## Pod Lifecycle Each iteration of Ralph: 1. **Select** next pending objective (by priority, then created_at) 2. **Mark** objective as "in_progress" 3. **Spawn** fresh Pod with objective context 4. **Pod executes:** - Worker implements - Inspector reviews requirements - Analyst reviews quality - Airlock validates (tests\/lint) 5. **Handle** Pod exit code: - 0: Mark complete, continue - 1: Mark failed, create blocker alert, try next - 2: Mark failed, create critical alert, HALT 6. **Repeat** until no pending objectives --- ## Alert Handling Ralph creates alerts automatically: | Situation | Severity | Ralph Action | |-----------|----------|--------------| | Pod exits 1 (blocker) | 1 | Create alert, continue to next objective | | Pod exits 2 (critical) | 0 | Create alert, HALT loop | | Max iterations reached | 0 | Create alert, HALT loop | | Existing critical alert | - | Check at start of each iteration, HALT if found | --- ## Safety Limits Ralph includes safety mechanisms: - **Max iterations:** 100 (prevents infinite loops) - **Critical alert check:** Before each objective - **Notification on halt:** macOS notification + notifications file --- ## Integration with CAPCOM During execution, Ralph logs to: - **Per-mission log:** `.space-agents\/missions\/active\/\/capcom.log` - **Notifications file:** `.space-agents\/comms\/notifications.md` Check progress with `\/capcom` - it reads these logs and SQLite state. --- ## Error Handling ### Mission not found ``` Ralph: Mission not found: MSN-999 Run \/mission-brief to create a mission first. ``` ### Mission not active ``` Ralph: Mission is not active (status: complete) Only active missions can be executed. ``` ### Database missing ``` Ralph: Database not found at .space-agents\/comms\/space-agents.db Run \/launch first to initialize Space-Agents. ``` ### Claude CLI not found ``` Ralph: claude CLI not found. Please install Claude Code CLI. ``` --- ## Example Session ``` User: \/mission-go HOUSTON: Active missions found: [1] MSN-001: JWT Token Management - 5 objectives pending Launching MSN-001 in attended mode... =========================================== RALPH LOOP STARTING =========================================== Mission: JWT Token Management (MSN-001) =========================================== --- Iteration 1 --- Selected objective: Implement token signing (OBJ-001) Spawning Pod... [Pod output streams here] Pod completed objective: Implement token signing Objective complete: OBJ-001 --- Iteration 2 --- ... ``` --- ## Troubleshooting ### Ralph hangs on first objective - Check if Claude CLI is properly installed - Verify the `agents\/mission-pod.md` file exists - Check SQLite permissions ### Objectives keep failing - Review alerts via `\/capcom` - Check the per-mission capcom.log for details - Consider running `\/airlock` manually to debug ### Critical alert halts everything - Run `\/capcom` to see the alert - Address the underlying issue - Clear the alert in SQLite - Re-run `\/mission-go` --- Ralph standing by. Ready to launch when you are.