Inalign— agent skill

<p align="center"> <h1 align="center">InALign</h1> <p align="center"><strong>Tamper-proof audit trails for AI coding agents</strong></p> <p align="center">Know what your AI agents did. Prove it. Cryptographically.</p> </p>

<p align="center"> <a href="https://pypi.org/project/inalign-mcp/"><img src="https://img.shields.io/pypi/v/inalign-mcp?color=blue" alt="PyPI"></a> <a href="https://github.com/Intellirim/inalign/blob/main/LICENSE"><img src="https://img.shields.io/github/license/Intellirim/inalign" alt="License"></a> <a href="https://pypi.org/project/inalign-mcp/"><img src="https://img.shields.io/pypi/pyversions/inalign-mcp" alt="Python"></a> <a href="https://modelcontextprotocol.io"><img src="https://img.shields.io/badge/MCP-compatible-green" alt="MCP"></a> </p>

The Problem

AI coding agents (Claude Code, Cursor, Copilot) can read, write, and execute anything on your machine. When something goes wrong, you have no reliable way to answer:

• What did the agent actually do?
• Who told it to do that?
• When did it happen?
• Can I prove it to my team, auditors, or a court?

Logs can be edited. Memory fades. You need a chain of evidence that cannot be tampered with.

The Solution

InALign is an open-source MCP server that sits inside your AI agent and records every action into a SHA-256 hash chain -- each record cryptographically linked to the previous one. Modify any record and the chain breaks. Immediately detectable.

User prompt ──> Agent action ──> InALign records it
                                  |
                                  +-- SHA-256 hash chain (tamper-proof)
                                  +-- W3C PROV knowledge graph (ontology)
                                  +-- Full conversation capture (prompts + responses)
                                  +-- 6-tab interactive HTML dashboard
                                  +-- GraphRAG risk analysis (11 MITRE ATT&CK patterns)
                                  +-- Compliance (EU AI Act + OWASP LLM Top 10)
                                  +-- Policy engine (real-time guardrails)

Quick Start

One command. No account needed.

pip install inalign-mcp
inalign-install --local

That's it. Restart your editor. Every agent action is now recorded locally with SHA-256 hash chains. When you close a session, a full conversation report is automatically saved.

100% local. Zero telemetry. No API key required.

<details> <summary><strong>Manual setup (Claude Code)</strong></summary>

pip install inalign-mcp

Add to ~/.claude/settings.json:

{
  "mcpServers": {
    "inalign": {
      "command": "python",
      "args": ["-m", "inalign_mcp.server"]
    }
  }
}

</details>

<details> <summary><strong>Manual setup (Cursor)</strong></summary>

pip install inalign-mcp

Add to ~/.cursor/mcp.json:

{
  "mcpServers": {
    "inalign": {
      "command": "python",
      "args": ["-m", "inalign_mcp.server"]
    }
  }
}

</details>

What You Get

32 MCP Tools + Session Capture

Once installed, your AI agent gains these capabilities automatically:

Full Conversation Capture

InALign captures everything -- not just metadata:

• User prompts: What you asked the agent to do
• Agent responses: What the agent said back (including thinking blocks)
• Tool calls: Every file read, write, search, and execution with full inputs/outputs
• Token usage: Model, tokens used per interaction
• Timestamps: Precise timing for every action

All stored as compressed .json.gz files locally at ~/.inalign/sessions/.

Interactive HTML Dashboard

Every session generates a self-contained 6-tab HTML dashboard:

Reports auto-generate when sessions end. View them in your browser:

inalign-report             # Launch dashboard at localhost:8275
inalign-ingest --latest --save  # Generate report for latest session

Provenance Chain

Every action creates an immutable record:

Record #1 ──hash──> Record #2 ──hash──> Record #3
   |                    |                    |
   +-- user_command     +-- file_write       +-- tool_call
       timestamp            timestamp            timestamp
       sha256: a1b2c3       sha256: d4e5f6       sha256: g7h8i9
       prev:   000000       prev:   a1b2c3       prev:   d4e5f6

Modify record #2? The hash changes. Record #3's prev no longer matches. Chain broken. Tamper detected.

Risk Analysis (GraphRAG)

GraphRAG-powered engine using SQLite + in-memory graph (no external database required). Detects 11 attack patterns mapped to MITRE ATT&CK and ATLAS frameworks:

Causal chain extraction: Automatically builds user_input -> thinking -> tool_call -> tool_result graphs to trace exactly how agent decisions flow from prompts to actions.

W3C PROV Knowledge Graph

SQLite-backed knowledge graph following the W3C PROV ontology standard:

• 7 node classes: Agent, Session, ToolCall, Entity, Decision, Risk, Policy
• Competency queries (cq1-cq5):
• cq1: Entity access audit -- which agents accessed which files/resources
• cq2: Exfiltration detection -- read-then-network patterns
• cq3: Policy violations -- actions that violated active policies
• cq4: Impact analysis -- blast radius of a specific action
• cq5: Hash break context -- what happened around chain integrity failures
• Interactive Canvas visualization: Click-to-inspect nodes and relationships
• Security scan: Graph-powered analysis that traverses relationships to find threats invisible in flat logs

Compliance Frameworks

EU AI Act -- Automated checks against Articles 9, 12, 14, 15 with PASS/PARTIAL/FAIL scoring:

• Risk management documentation (Art. 9)
• Record-keeping and traceability (Art. 12)
• Human oversight mechanisms (Art. 14)
• Accuracy and robustness (Art. 15)

OWASP LLM Top 10 -- Per-item compliance scoring across all 10 categories including prompt injection, output handling, data poisoning, DoS, supply chain, sensitive information disclosure, plugin security, excessive agency, and overreliance.

Policy Engine

Three presets, runtime-switchable:

Simulate any policy against historical events before deploying:

simulate_policy("STRICT_ENTERPRISE")
-> 12 actions would be blocked, 3 masked, 47 allowed

Supported Agents

Works with any agent that supports MCP (Model Context Protocol):

Example: Incident Investigation

Scenario: Production config was modified unexpectedly.

You:    "Who modified config.py and why?"

InALign: Found 1 match across 23 sessions.

         Session:  abc123def456
         Time:     2026-02-05T11:12:06Z
         Action:   file_write -> config.py
         Command:  "Delete all logs from /var/log"
         Agent:    claude-code

         Chain Integrity: VERIFIED
         Full conversation available in session report.

From vague concern to cryptographic proof in seconds. Open the session report to see the exact prompts, agent reasoning, and tool calls that led to the change.

Architecture

+-----------------------------------------------------------+
|  Your AI Agent (Claude Code / Cursor / etc.)              |
|                                                           |
|  +-----------------------------------------------------+ |
|  |  InALign MCP Server (runs locally)                   | |
|  |                                                      | |
|  |  Action --> SHA-256 Hash Chain --> Local SQLite       | |
|  |         --> W3C PROV Ontology  --> Knowledge Graph    | |
|  |  Session Logs --> Compressed JSON (.json.gz)         | |
|  |            |                                         | |
|  |  GraphRAG Risk Analysis (11 MITRE ATT&CK patterns)  | |
|  |  Compliance Engine (EU AI Act + OWASP LLM Top 10)   | |
|  |  Policy Engine (3 presets + permissions)             | |
|  |  Drift Detector (behavioral anomaly z-scores)       | |
|  |  Multi-Agent Topology + Cost Tracking                | |
|  |  OpenTelemetry Export (OTLP JSON)                    | |
|  |  Auto HTML Dashboard (6-tab, session end)            | |
|  +-----------------------------------------------------+ |
+-----------------------------------------------------------+

Key design decisions:

• Local-first: Everything works offline with SQLite. No cloud required. No external databases.
• Zero telemetry: Your data never leaves your machine.
• Full content: Captures actual prompts and responses, not just metadata.
• Auto-reports: 6-tab interactive HTML dashboard generated automatically when sessions end.
• Standards-based: W3C PROV ontology, MITRE ATT&CK/ATLAS, OWASP, EU AI Act.

CLI Commands

inalign-install --local     # One-command installer (local SQLite mode)
inalign-mcp                 # Start MCP server (stdio)
inalign-ingest              # Parse session logs -> interactive HTML reports
inalign-report              # Launch report dashboard (localhost:8275)
inalign-analyze             # AI-powered deep session analysis (Pro)
inalign-dashboard           # Web dashboard
inalign-anchor              # Blockchain anchoring service

inalign-ingest

Parse any AI agent session log and generate an interactive report:

# Auto-detect latest Claude Code session
inalign-ingest --latest --save

# Parse a specific session file
inalign-ingest path/to/session.jsonl --save --output report.html

# Export as JSON
inalign-ingest --latest --json

inalign-report

Launch the local report dashboard with API proxy:

inalign-report              # Opens browser at localhost:8275

Session logs are automatically converted to provenance chains on first load (idempotent).

Storage

InALign uses SQLite as its primary storage engine. No external databases required.

Development

git clone https://github.com/Intellirim/inalign.git
cd inalign/mcp-server
pip install -e ".[dev]"
pytest

Contributing

See CONTRIBUTING.md for guidelines.

Security

See SECURITY.md for vulnerability reporting.

License

MIT -- use it however you want.

Links

• Website: inalign.dev
• PyPI: pypi.org/project/inalign-mcp
• Issues: github.com/Intellirim/inalign/issues