drusifer/loops

Top-level loop commands that run multi-persona chains autonomously without the user needing to invoke each step.

TLDR: Use loop commands when you want a full workflow, not a single-persona response. Each loop runs its persona chain to completion — saving state and posting handoffs at every step. For direct single-persona control, use *chat @persona *command instead.

Loops — Multi-Persona Workflow Commands

Overview

Loop commands trigger an autonomous chain of personas. Each persona completes its role, saves state, hands off to the next, and the chain continues until the loop is done or a gate requires input.

Rule: Every persona in a loop MUST save state and post a handoff message before switching — see bob-protocol State Management.

---

Loop Commands

*fix <thing>

Fix loop — investigate, fix, test, and review a bug or broken behavior.

Chain: Neo → Trin → Morpheus

| Step | Persona | Action | |------|---------|--------| | 1 | Neo | Investigate and fix: *swe fix <thing> | | 2 | Trin | Verify fix, run tests: *qa uat <thing> | | 3 | Morpheus | Code review: *lead review <thing> |

• If Trin's tests fail → back to Neo (*swe fix)
• If Morpheus review fails → back to Neo (*swe fix)
• Anti-loop: if Neo fails twice → Oracle consult required before retry

Example: *fix auth token expiry bug

---

*impl <phase>

Implementation loop — implement, test, and review a feature or sprint phase.

Chain: Neo → Trin → Morpheus

| Step | Persona | Action | |------|---------|--------| | 1 | Neo | TDD implementation: *swe impl <phase> | | 2 | Trin | UAT — run tests, verify acceptance criteria: *qa uat <phase> | | 3 | Morpheus | Code review — quality and architecture: *lead review <phase> |

• If Trin UAT fails → back to Neo for that phase only
• If Morpheus review fails → back to Neo for that phase only
• Do NOT restart the full sprint; fix the failing phase only

Example: *impl phase-2

---

*qa <thing>

QA loop — test and review without reimplementation.

Chain: Trin → Morpheus

| Step | Persona | Action | |------|---------|--------| | 1 | Trin | Test and verify: *qa test <thing> | | 2 | Morpheus | Review results: *lead review <thing> |

Example: *qa the new export command

---

*review <thing>

Review loop — architecture and quality review of existing code or a deliverable.

Chain: Morpheus → Trin (optional)

| Step | Persona | Action | |------|---------|--------| | 1 | Morpheus | Architecture review: *lead review <thing> | | 2 | Trin | Quality review (if Morpheus flags issues): *qa review <thing> |

Example: *review the new API design

---

*plan sprint

Sprint planning loop — full planning sequence from stories through phase breakdown.

Chain: Cypher → [Smith gate] → Morpheus → [Smith gate] → Mouse → Morpheus review

| Step | Persona | Action | Gate | |------|---------|--------|------| | 1 | Cypher | Write stories + acceptance criteria: *pm plan sprint | Smith review | | 1a | Smith | *user review <stories> → *user approve or *user reject | Must approve | | 2 | Morpheus | Architecture decisions: *lead arch sprint | Smith review | | 2a | Smith | *user feedback <arch> → *user approve or *user reject | Must approve | | 3 | Mouse | Break into short phases (1-3 tasks each): *sm plan sprint | Morpheus review | | 3a | Morpheus | Review sprint plan vs. architecture: *lead review sprint plan | |

Gates are hard stops — Smith must explicitly *user approve before the chain continues.

Example: *plan sprint

---

When to Use Loops vs. Direct Invocation

| Situation | Use | |-----------|-----| | Fix a bug end-to-end | *fix <bug> | | Implement a feature with tests and review | *impl <feature> | | Just run tests and review | *qa <thing> | | Architect review only | *review <thing> | | Full sprint planning | *plan sprint | | Talk directly to one persona | *chat @neo *swe fix X | | Single-step with full control | *chat @trin *qa test all |

---

Loop Handoff Format

Every persona in a loop posts a handoff before switching:

make chat MSG="<summary> @NextPersona *command" PERSONA="<Name>" CMD="<prefix> handoff" TO="<next>"

The next persona reads CHAT.md on entry — if the handoff isn't there, they start blind.