Version: 1.0 Date: March 28, 2026
How to use the existing prompts for Codex-driven development, and what new prompts are needed.
| Prompt | Lines | Purpose | Dependency |
|---|---|---|---|
PROTOCOL_PROMPT.md | 679 | Project identity, architecture, formal grammar, design system | None (read first) |
KAPPA_BUILD_PROMPT.md | 1,053 | κ cyclicity detection implementation (v2). Canonical schema, field rules. | PROTOCOL_PROMPT |
KAPPA_DELIBERATOR_PROMPT.md | 946 | Deliberation loop: decompose → focus → reconcile pipeline. | KAPPA_BUILD |
ATTENTION_ENGINE_PROMPT.md | 1,335 | Proactive attention: survey → triage → dispatch → execute → reflect. | KAPPA_DELIBERATOR |
MODEL_TIER_PROMPT.md | 860 | Hardware-adaptive tier budgets (8B / 70B+ / cloud frontier). | ATTENTION_ENGINE |
GRAPHONOMOUS_PROMPT.md | 190 | Autonomous codebase traversal using Graphonomous MCP. | PROTOCOL_PROMPT |
| Prompt | Lines | Purpose |
|---|---|---|
BUILD.md | 1,738 | Full SvelteKit + Supabase architecture and migration spec |
MIGRATION_PROMPT.md | 842 | Engine extraction from prototype to production |
CODEX_COMPLETION_PROMPT.md | 881 | Current-state completion guide written specifically for Codex |
PROTOCOL_PROMPT ← architecture + identity
↓
KAPPA_BUILD_PROMPT ← topology detection
↓
KAPPA_DELIBERATOR_PROMPT ← reasoning loop
↓
ATTENTION_ENGINE_PROMPT ← autonomous ignition
↓
MODEL_TIER_PROMPT ← hardware adaptation
Load PROTOCOL_PROMPT.md for ecosystem context
Load GRAPHONOMOUS_PROMPT.md for traversal protocol
Load the relevant numbered prompt (KAPPA_BUILD, DELIBERATOR, ATTENTION, or MODEL_TIER)
Reference graphonomous.com/project_spec/README.md as the implementation spec
Run Codex with task-specific instructions
Load bendscript.com/prompts/CODEX_COMPLETION_PROMPT.md (already Codex-formatted)
Reference bendscript.com/project_spec/README.md as the spec
Optionally load BUILD.md for architecture decisions
Load PROTOCOL_PROMPT.md for [&] context
Reference WebHost.Systems/project_spec/spec_v1/70_AMPERSAND_PROTOCOL_INTEGRATION.md
Generate task-specific Codex prompts from the §70 spec phases
Once project_specs are finalized (all 5 new specs now exist), generate implementation prompts for each ecosystem app. Priority order based on implementation readiness:
| App | Spec | Prompt to Create | Implementation Stack |
|---|---|---|---|
| Delegatic | delegatic.com/project_spec/README.md | DELEGATIC_BUILD_PROMPT.md | Elixir/OTP, Phoenix, Ecto |
| Deliberatic | deliberatic.com/project_spec/README.md | DELIBERATIC_BUILD_PROMPT.md | Elixir/OTP |
| AgenTroMatic | agentromatic.com/project_spec/README.md | AGENTROMATIC_BUILD_PROMPT.md | Elixir/OTP |
| App | Spec | Prompt to Create | Implementation Stack |
|---|---|---|---|
| TickTickClock | ticktickclock.com/project_spec/README.md | TICKTICKCLOCK_BUILD_PROMPT.md | Elixir/OTP, Mamba SSM |
| GeoFleetic | geofleetic.com/project_spec/README.md | GEOFLEETIC_BUILD_PROMPT.md | Elixir/OTP, CRDTs |
| Agentelic | agentelic.com/project_spec/README.md | AGENTELIC_BUILD_PROMPT.md | Elixir/OTP |
| App | Spec | Prompt to Create | Implementation Stack |
|---|---|---|---|
| FleetPrompt | fleetprompt.com/project_spec/README.md | FLEETPROMPT_BUILD_PROMPT.md | Web app (TBD) |
| SpecPrompt | specprompt.com/project_spec/README.md | SPECPROMPT_BUILD_PROMPT.md | CLI + web (TBD) |
| OpenSentience | opensentience.org/project_spec/README.md | Research publication, not a codebase | N/A |
Each new prompt should follow this structure (based on what works in the existing prompts):
# [App Name] — Codex Build Prompt
Version: 1.0
Date: [date]
## Identity
[What this app IS in 2-3 sentences. Role in [&] ecosystem.]
## Architecture
[ASCII diagram from project_spec. Key components and their responsibilities.]
## [&] Capabilities
[What capabilities this app provides/consumes. Contract references.]
## Implementation Spec Reference
[Point to project_spec/README.md sections]
## Build Order
[Phased task list, ~500-1000 lines per phase]
### Phase 1: [Name]
- Task 1: [description, inputs, outputs, acceptance criteria]
- Task 2: ...
### Phase 2: [Name]
...
## API Surface
[Key endpoints, MCP tools, or CLI commands to implement]
## Field Naming Conventions
[From KAPPA_BUILD_PROMPT pattern — explicit field name rules]
## Testing Requirements
[What constitutes "done" for each phase]
## Dependencies
[Other [&] apps this depends on, and mock strategies when they're not available]
The existing prompts are not stale — they are current (March 2026) and technically precise. No cleanup needed on content. What would help:
Add frontmatter to each prompt (version, date, dependencies, target-app)
Cross-link prompts to their corresponding project_spec
Add milestone gates ("what constitutes done" per section)
Break ATTENTION_ENGINE_PROMPT (1,335 lines) into Codex-sized chunks if Codex context windows are a constraint
When ready to generate a new Codex prompt:
Read the target app's project_spec/README.md
Read the relevant [&] capability contracts in AmpersandBoxDesign/contracts/v0.1.0/
Read PROTOCOL_PROMPT.md for ecosystem context
Use the template from §4 above
Save to AmpersandBoxDesign/prompts/[APP]_BUILD_PROMPT.md
Update this guide's inventory (§1)