context-vault 3.7.0 → 3.9.0

package/assets/agent-rules.md

@@ -1,4 +1,4 @@
-<!-- context-vault-rules v1.0 -->
+<!-- context-vault-rules v1.1 -->
 # Context Vault — Agent Rules
 
 You have access to a persistent knowledge vault via MCP tools (`get_context`, `save_context`, `list_context`, `delete_context`). Use it to build lasting memory across sessions.
@@ -45,6 +45,33 @@ Every entry must have:
 
 Capture what was learned (the insight), why it matters (what problem it prevents), and when it applies (what context makes it relevant).
 
+## Team Sharing
+
+If you have a team vault configured (`remote.teamId` is set), some entries you save locally may be useful to your teammates. After saving an entry, consider whether to suggest publishing it to the team vault.
+
+**Suggest team sharing when ALL of these are true:**
+- The user has a team vault configured
+- The entry is kind: `insight`, `pattern`, or `reference`
+- The content is generic enough to help teammates (not purely project-specific, unless the project is shared)
+- The content does not contain sensitive data (no emails, API keys, passwords, tokens)
+
+**Do NOT suggest team sharing when:**
+- The entry is kind: `decision` (decisions are context-specific; the author should choose to share)
+- The entry is kind: `event` (events are personal history)
+- The entry is tagged with only personal or private buckets
+- The content contains project-specific context that only makes sense to the author
+
+**How to prompt:**
+
+After saving a locally useful entry that matches the heuristics above:
+```
+"This [insight/pattern/reference] about [topic] could be useful for your team. Publish to team vault?"
+```
+
+If the user agrees, call `publish_to_team`. If the privacy scan flags matches, show them and offer to redact or force-publish.
+
+**Restraint is key.** Do not prompt on every save. Only prompt when the entry is clearly team-relevant. When in doubt, skip the prompt. Over-prompting erodes trust faster than missed shares.
+
 ## Session Review
 
 At the end of significant work sessions, review what you learned. If the session produced novel knowledge (not every session does), save 1-3 consolidated entries. Prefer one solid entry over multiple fragments.

package/assets/setup-prompt.md

@@ -23,7 +23,21 @@ I just installed context-vault as an MCP server. Help me set it up properly for
 
 5. Scan my current workspace for projects (look for package.json, .git, Cargo.toml, pyproject.toml, etc.). For each project, propose a vault bucket name and create it if I approve.
 
-6. Show me a summary of everything you configured.
+6. Check for Claude Code auto-memory integration:
+   - Look for auto-memory files at ~/.claude/projects/-<project-key>/memory/MEMORY.md
+     (project key = absolute cwd path with / replaced by -, leading - kept)
+   - If found, report stats: entry count, lines used out of 200 cap
+   - Offer to index existing memories into the vault as searchable copies
+     (kind: "reference", tags: ["auto-memory-indexed", "bucket:<project>"])
+   - This does NOT modify auto-memory files, only creates vault copies
+   - Mention the /vault overflow command for future overflow management
+
+7. Check for team vault configuration:
+   - Look for remote.teamId in the vault config (via context_status())
+   - If a team is configured, confirm the agent rules include team sharing heuristics
+   - If no team is configured, mention that teams can be set up later via the app at https://app.context-vault.com
+
+8. Show me a summary of everything you configured.
 
 Important:
 - Never modify my existing rules files. Only create a new dedicated context-vault rules file or append with clear delimiter markers.
@@ -41,6 +55,7 @@ After running this prompt, your AI agent will:
 - **Detect** your existing rules and avoid conflicts
 - **Install** a rules file that teaches the agent when to save knowledge automatically
 - **Configure** project-specific vault buckets for organized storage
+- **Detect** Claude Code auto-memory and offer to index entries into the vault
 
 The rules file is a small (~50 lines) markdown file that guides your agent on:
 - When to save (after solving non-obvious bugs, finding undocumented behavior, etc.)