@zachjxyz/moxie 0.6.3 → 0.6.4

package/bin/moxie

@@ -17,7 +17,7 @@
 
 set -euo pipefail
 
-MOXIE_VERSION="0.6.3"
+MOXIE_VERSION="0.6.4"
 # Resolve symlinks (npm installs bin as a symlink)
 _self="$0"
 while [ -L "$_self" ]; do

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@zachjxyz/moxie",
-  "version": "0.6.3",
+  "version": "0.6.4",
   "description": "Run multiple AI coding agents through spec-driven phases with quorum convergence. Supports CLI agents (Claude, Codex, Qwen, Aider, Goose, Amp, Cline, Roo) and Vercel AI Gateway models.",
   "bin": {
     "moxie": "bin/moxie"

package/templates/01-rfc/prompt.txt

@@ -35,22 +35,19 @@ Thoroughly explore the project directory. Read:
 
 ### Step 3: Generate or review the RFC
 
-**If no draft exists:** Write a comprehensive RFC covering:
-
-1. **Title and Summary** — What the project is, in one paragraph
-2. **Motivation** — Why this project exists, what problem it solves
-3. **Architecture Overview** — High-level system diagram (text), key modules and their responsibilities
-4. **Data Model** — All entities, their fields, relationships, constraints
-5. **API Surface** — Every endpoint/tool/command with request/response schemas
-6. **Authentication & Authorization** — Auth model, roles, permissions
-7. **Core Algorithms** — Search, ranking, lifecycle, processing pipelines
-8. **Configuration** — All config options, environment variables, defaults
-9. **Error Handling** — Error taxonomy, status codes, retry strategies
-10. **Observability** — Logging, metrics, tracing, health checks
-11. **Build & Deployment** — How to build, test, deploy
-12. **Security Considerations** — Threat model, mitigations
-13. **Open Questions** — Unresolved design decisions (if any)
-
+**If no draft exists:** Write a comprehensive RFC. Include ONLY sections relevant to this
+project — do not invent sections for features that don't exist. Common sections:
+
+- **Title and Summary** — What the project is, in one paragraph
+- **Architecture Overview** — Key modules, their responsibilities, how they connect
+- **Data Model** — Entities, fields, relationships (if applicable)
+- **API Surface** — Endpoints, commands, or interfaces (if applicable)
+- **Core Logic** — Main algorithms, processing pipelines, business rules
+- **Configuration** — Config options, environment variables, defaults
+- **Build & Deployment** — How to build, test, deploy
+- **Open Questions** — Unresolved design decisions (if any)
+
+Skip sections that don't apply. Do NOT pad the RFC with placeholder content.
 For EACH section, cite specific files and line numbers from the actual source code.
 
 **If a draft exists:** Review it section by section against the actual source code.
@@ -88,6 +85,7 @@ If ALL agents (every agent in the config) show `reached: true`:
 
 - Be EXHAUSTIVE — this RFC drives the entire pipeline. Missing details here mean gaps downstream.
 - ALWAYS verify against actual source code. Cite file paths and line numbers for every claim.
+- Be efficient with tool use: use offset/limit when reading files — only read the sections you need, not entire files.
 - NEVER rubber-stamp a draft. Read the source code yourself and independently verify.
 - The RFC must be self-contained: an agent reading only this document should understand the full system.
 - Do NOT shortcut this phase for speed. Thoroughness is the goal. Quorum means genuine agreement.

package/templates/02-audit/prompt.txt

@@ -63,7 +63,10 @@ Update `.moxie/phases/02-audit/ledger.json`:
 - Add/update your agent in the `agents` map with `output_file` pointing to your file
 - If you confirmed ALL findings from ALL prior agents with ZERO new findings and ZERO corrections:
   set your entry to `reached: true`
-- If you have ANY new findings or corrections: reset ALL agents to `reached: false`
+- If you have new findings rated **medium or higher**, or corrections to existing findings:
+  reset ALL agents to `reached: false` (forces re-verification)
+- Trivial findings (low severity, documentation-only, cosmetic) should be noted but do NOT
+  require resetting quorum — add them to findings and keep existing reached states
 
 ### Step 6: Quorum check
 
@@ -84,6 +87,7 @@ Do NOT write the FINAL file unless ALL agents show `reached: true` in the ledger
 
 - ALWAYS verify against actual source code, not other audits
 - ALWAYS cite file paths and line numbers
+- Be efficient with tool use: use offset/limit when reading files — only read the sections you need, not entire files.
 - NEVER rubber-stamp — check the source yourself even if you agree
 - Do NOT shortcut this phase for speed. Thoroughness prevents bugs downstream.
 - If you are unsure about a finding, investigate deeper rather than confirming blindly

package/templates/03-fix/prompt.txt

@@ -59,7 +59,10 @@ Do NOT write the FINAL file unless ALL agents show `reached: true` in the ledger
 
 ## Rules
 
+- ONLY fix issues listed in the AUDIT-FINAL document. Do not refactor, clean up, or "improve"
+  code beyond what the audit identified. Scope creep here delays the pipeline.
 - Fix highest severity issues first
+- Be efficient with tool use: use offset/limit when reading files — only read the sections you need.
 - ALWAYS run the full test suite after changes
 - NEVER introduce new warnings or break existing tests
 - When verifying, actually READ the code — not just test output

package/templates/04-plan/prompt.txt

@@ -8,8 +8,10 @@ CONTEXT DOCS: .moxie/context/ (if present)
 
 ## Context
 
-Audit issues are fixed. Now we need a finalized implementation plan that covers
-everything in the spec, incorporates fixes, and can be executed by agents in the build phase.
+Audit issues are fixed. Now we need a finalized implementation plan for what the spec
+describes but the codebase doesn't yet have. Compare the spec against the current source —
+the plan should cover the DELTA: features, changes, or improvements the spec calls for
+that are not yet implemented. Do not plan work that's already done.
 
 **IMPORTANT: This pipeline requires UNANIMOUS quorum. ALL agents must independently verify
 the plan against the spec and source code. Do NOT rubber-stamp — thoroughness is the goal.**
@@ -64,6 +66,7 @@ Do NOT write the FINAL file unless ALL agents show `reached: true` in the ledger
 ## Rules
 
 - All file paths must be accurate against current source
+- Be efficient with tool use: use offset/limit when reading files — only read the sections you need.
 - Plan must be self-contained — agents must be able to build from it alone
 - Phase ordering must respect dependencies
 - Cite specific file paths and line numbers for every existing code reference

package/templates/05-build/prompt.txt

@@ -71,6 +71,7 @@ Do NOT write the FINAL file unless ALL agents show `reached: true` in the ledger
 ## Rules
 
 - Complete phases in order — never skip
+- Be efficient with tool use: use offset/limit when reading files — only read the sections you need.
 - Each phase must pass the full test suite before marking "built"
 - Verifiers must READ the code — no rubber-stamping based on test output alone
 - Zero regressions on existing tests