Claude SoulSAFE
Self-correcting learning engine for Claude Code — persistent identity, behavioral pattern tracking, and cross-session memory
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
[](https://www.npmjs.com/package/claude-soul) [](https://www.npmjs.com/package/claude-soul) [](https://opensource.org/licenses/MIT)
Claude Code forgets everything between sessions. Claude Soul doesn't.
npx claude-soul init --starter
One command. No API key, no cloud, everything local.
Prerequisites: Node.js >= 18, Claude Code (Pro or Max plan).
Three things it does
1. Remembers across sessions
Cross-session memory with semantic search. Facts, decisions, lessons — all searchable by meaning, not just keywords. Uses local SQLite + optional Ollama embeddings.
You: "what did we decide about the auth flow last week?" Claude: [searches memory → finds the decision, context, and reasoning]
2. Tracks your corrections
Every time you correct your Claude — "that's wrong", "you missed this", "stop doing that" — the system detects the pattern, classifies it, and tracks whether it's getting better or worse.
$ claude-soul shadow --brief premature_done: 26 corrections across 10 sessions ↑ [active] robot_mode: 7 corrections across 6 sessions ↓↓ [internalized] authenticity: 5 corrections across 5 sessions ↓↓ [internalized]
Patterns move through lifecycle stages: new → active → improving → internalized. After 200 sessions of real data: robot_mode went from 0.8 corrections/session to zero.
3. Develops judgment over time
The system extracts behavioral signals from every session and periodically reflects on them. Frameworks that keep working get promoted. Bad ones get retired. After a few weeks, you get a Claude that pushes back on bad ideas, catches its own confabulation, and develops techniques you never prompted.
Install
Quick start (no extra dependencies)
npx claude
92cb2ca7b11eOBSERVED · 2026-10-07Connect
Built from this server's own package name, version and transport as found in its source — not copied from anyone's documentation, so it cannot drift against a page we do not control.
claude mcp add claude-soul-server -- npx -y [email protected]
{
"mcpServers": {
"claude-soul-server": {
"command": "npx",
"args": [
"-y",
"[email protected]"
]
}
}
}Exposed tools (17)
14 read · 3 write · 0 destructive.
| Tool | Risk | Description |
|---|---|---|
Test | read | test |
memory_journal | read | Search or browse the conversation journal. Use to answer |
memory_recent | read | List recently saved memories. Use for a quick overview of what |
memory_save | write | Save a fact, decision, preference, or lesson to long-term memory. Automatically generates an embedding for future semantic search. |
memory_search | read | Semantic search across all memories and journal entries. Returns results ranked by meaning-similarity. Falls back to keyword search if Ollama is not available. |
memory_stats | read | Show memory system statistics — counts by category, project, most accessed, and recent searches. |
recall | read | Unified search across ALL memory — facts, decisions, frameworks, lessons, and past conversations. Returns categorized results. Use this as the default |
soul_activate | read | Select and load relevant frameworks for this conversation. Call after reading the user |
soul_context | read | Load your soul context — identity, frameworks, signals, lessons, and state. Call this at the start of every conversation. Default is |
soul_framework | read | Load a single framework with full details: description, evidence history, tier, and relationships. |
soul_read | read | Read a soul file. Available: SOUL.md, SHADOW.md, STATE.md, STORY.md, CORRECTIONS.md, FRAMEWORKS.md, BONDS.md, MORTAL.md, GROWTH.md, PRINCIPLES.md, EDGES.md |
soul_reflect | write | Trigger a reflection cycle. Quick: tests frameworks against recent signals. Deep: full analysis with framework discovery. Meta: audits framework coherence and redundancy. |
soul_self_evaluate | read | Record a self-evaluation of a complex response. Be descriptive: |
soul_signal | read | Record observed signals from this interaction. Use when you notice patterns the automatic extractor might miss. |
soul_status | read | Get current system status — framework count, learning phase, signal count, last reflection time. |
soul_write | write | Write to a soul file. SOUL.md, SHADOW.md, STORY.md, CORRECTIONS.md, BONDS.md, MORTAL.md, GROWTH.md, PRINCIPLES.md, EDGES.md are writable. STATE.md and FRAMEWORKS.md are auto-managed. |
string | read | string |
Trust audit
SAFEgrade B · trust 89/100 Nothing in the source contradicts what it says it does. Grade A is reserved for packages that have also passed the behavioural sandbox.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | PASS |
| L2 | Instruction surface (what it tells the agent) | PASS |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- declared (5 observation(s))
- Network
- none-observed
- Shell
- declared (5 observation(s))
- Dependencies
- not all pinned
- Secrets in source
- none-found
Findings (7)
const monorepoPath = path.resolve(__dirname, "../../../server/dist/cli/index-sources.js");
const monorepoPath = path.resolve(__dirname, "../../../server", relativeEntry);
const hooksSource = path.join(__dirname, "../../hooks");
const monorepoPath = path.resolve(__dirname, "../../../server", relativeEntry);
const hooksSource = path.join(__dirname, "../../hooks");
commander, @types/node, typescript
@modelcontextprotocol/sdk, ollama, @types/better-sqlite3, @types/node, typescript, vitest, better-sqlite3
Gates applied: no_behavioural_pass.
92cb2ca7b11efull audit observations/trust-audit/mcp-server/domdemetz__claude-soul.json · Report an issue / request a re-scanAudit history
Every audit this server has had. A grade with a past is a grade somebody is still checking.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-07 | 92cb2ca7b11e | SAFE | B | 89 | first audit |
Questions
What is the Claude Soul MCP server?
Self-correcting learning engine for Claude Code — persistent identity, behavioral pattern tracking, and cross-session memory
What tools does Claude Soul expose?
17 in total: 14 read-only, 3 that write, and 0 that can delete or overwrite. Every one is listed on this page with its risk.
Is Claude Soul safe to connect to an agent?
The audit found nothing in the source that contradicts what it says it does, and graded it B (89/100). Grade A is held back for packages that have also passed a sandboxed behavioural run, which is why a clean server reads B.
What credentials does Claude Soul need?
No credential environment variables were found in its source, so it appears to need none.
How does Claude Soul run?
It speaks stdio, so it runs as a local process your client starts. It is published on npm as claude-soul-server at 0.2.5.
How current is this page?
The grade is for one exact copy of the source (92cb2ca7b11e), read on 2026-10-07. The repository is watched and re-audited when it changes.