claude-code-autoconfig 1.0.118 → 1.0.120

package/.claude/commands/autoconfig.md

@@ -1,5 +1,5 @@
 <!-- @description The command you just ran. Analyzes your project and populates CLAUDE.md with real context. Re-run anytime your stack changes. -->
-<!-- @version 2 -->
+<!-- @version 3 -->
 
 # Autoconfig
 
@@ -103,12 +103,12 @@ Focus on what Claude Code actually needs to work effectively. Claude can explore
 **Wrap content in markers** so `/sync-claude-md` knows what's auto-generated:
 
 ```markdown
-<!-- AUTO-GENERATED BY /autoconfig at {TIMESTAMP} UTC — Do not edit. Use .claude/feedback/ for corrections. -->
+<!-- AUTO-GENERATED BY /autoconfig at {TIMESTAMP} UTC — Do not edit between markers. -->
 
 # Project Name
 ...content...
 
-<!-- END AUTO-GENERATED at {TIMESTAMP} UTC — Use .claude/feedback/ for corrections. -->
+<!-- END AUTO-GENERATED at {TIMESTAMP} UTC -->
 ```
 
 Replace `{TIMESTAMP}` with the current UTC time in format `YYYY-MM-DD HH:MM:SS` (e.g., `2026-01-14 16:30:45`). Use the same timestamp in both markers.
@@ -148,14 +148,27 @@ Place this section near the top (after Tech Stack, before Commands) since versio
 
 **Keep it tight.** A 30-line CLAUDE.md that hits the essentials beats a 200-line doc Claude has to parse every session.
 
-**Always end with:**
+**Always end with (inside markers):**
 ```markdown
 ## Team Feedback
 The contents of `.claude/feedback/FEEDBACK.md` are an extension of this file.
 Read it at the start of every session before taking any action.
+FEEDBACK.md is reserved for human-authored corrections only — do not write to it.
 ```
 
-This pointer persists across autoconfig runs and ensures FEEDBACK.md guidance is reliably loaded every session.
+**Always add below the end marker:**
+```markdown
+
+## Discoveries
+<!-- Claude: append project-specific learnings, gotchas, and context below. This section persists across /autoconfig runs. -->
+
+```
+
+This creates two zones in CLAUDE.md:
+1. **Between markers** — auto-generated by /autoconfig, never manually edited
+2. **Below markers (Discoveries section)** — Claude appends learnings here instead of FEEDBACK.md
+
+When Claude discovers project-specific context (related projects, non-obvious gotchas, critical workflows) during a session, it should append to the Discoveries section. This keeps FEEDBACK.md clean for human team corrections only.
 
 ## Step 4: Create Rules Directory
 

package/.claude/commands/sync-claude-md.md

@@ -1,5 +1,5 @@
 <!-- @description Re-analyzes your project and updates CLAUDE.md to reflect current state. -->
-<!-- @version 1 -->
+<!-- @version 2 -->
 
 # Sync CLAUDE.md
 
@@ -45,14 +45,14 @@ If multiple sources exist with different values, include a `## Version Managemen
 CLAUDE.md uses markers to identify auto-generated content:
 
 ```markdown
-<!-- AUTO-GENERATED BY /autoconfig at {TIMESTAMP} UTC — Do not edit. Use .claude/feedback/ for corrections. -->
+<!-- AUTO-GENERATED BY /autoconfig at {TIMESTAMP} UTC — Do not edit between markers. -->
 ...content...
-<!-- END AUTO-GENERATED at {TIMESTAMP} UTC — Use .claude/feedback/ for corrections. -->
+<!-- END AUTO-GENERATED at {TIMESTAMP} UTC -->
 ```
 
 **IMPORTANT:** Always replace `{TIMESTAMP}` with the CURRENT UTC time in format `YYYY-MM-DD HH:MM:SS` (e.g., `2026-01-14 16:30:45`).
 
-**Only regenerate content between these markers.** Content outside markers will be removed on next sync - users should add feedback to `.claude/feedback/FEEDBACK.md` instead.
+**Only regenerate content between these markers.** Content below the end marker (e.g., the `## Discoveries` section) is Claude-authored and must be preserved across syncs.
 
 Compare detected state with current CLAUDE.md and update:
 
@@ -78,6 +78,7 @@ Show the user what changed in CLAUDE.md before finishing.
 ## Guidelines
 
 - **Respect markers**: Only regenerate content between AUTO-GENERATED markers
+- **Preserve Discoveries**: Content below the end marker (## Discoveries section) is Claude-authored and must never be removed or overwritten during sync
 - **Be conservative**: Only update sections that actually changed
 - **Keep it tight**: A 30-line CLAUDE.md that hits the essentials beats a 200-line doc Claude has to parse every session
 - **No boilerplate**: If Claude can infer it from scanning `package.json` or config files, don't put it in CLAUDE.md

package/.claude/feedback/FEEDBACK.md

@@ -1,9 +1,9 @@
-<!-- @description Team-maintained corrections and guidance for Claude. Add notes here when Claude does something wrong — it learns for next time. This directory persists across /autoconfig runs. -->
+<!-- @description Human-authored corrections and guidance for Claude. Reserved for team feedback only — Claude must not write here. This directory persists across /autoconfig runs. -->
 
 # Team Feedback
 
-Add corrections and guidance here when Claude does something wrong.
-Claude reads this directory and learns for next time.
+**This file is for human-authored corrections and guidance only.**
+Claude reads this file but must never write to it. When Claude discovers project context, gotchas, or learnings, it should append to the `## Discoveries` section in CLAUDE.md instead.
 
 ---
 

package/.claude/settings.json

@@ -58,8 +58,6 @@
       "Read(./**/credentials.*)",
       "Read(./**/*.pem)",
       "Read(./**/*.key)",
-      "Write(./nul)",
-      "Edit(./nul)",
       "Bash(rm -rf:*)",
       "Bash(curl:*)",
       "Bash(wget:*)"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "claude-code-autoconfig",
-  "version": "1.0.118",
+  "version": "1.0.120",
   "description": "Intelligent, self-configuring setup for Claude Code. One command analyzes your project, configures Claude, and shows you what it did.",
   "author": "ADAC 1001 <info@adac1001.com>",
   "license": "MIT",