IDRIS Documentation

Overview

IDRIS is not an offensive tool. It's the discovery and governance layer the entire Red Specter stack was missing. It operates across six core modules:

• Agent Discovery Engine — 10 sources, continuous scanning
• Permission Mapper — escalation path detection, orphaned credential tracking
• Ownership Tracer — accountability, departed owner detection
• Identity Graph — relationship mapping, blast radius, trust chains
• Compliance Audit Generator — 5 frameworks, gap analysis
• NEMESIS Integration Feed — adversarial validation loop

The full lifecycle: IDRIS discovers. NEMESIS validates. AI Shield defends.

Module 1: Agent Discovery Engine

Continuously scans across 10 discovery sources to find every AI agent deployed in your environment.

Module 2: Permission Mapper

Traces every permission grant for every discovered agent and identifies escalation paths.

For each agent, the Permission Mapper:

• Counts permissions by scope: admin, write, read, execute, delegate
• Identifies orphaned credentials (resource deleted but key still active)
• Detects expired permissions still present in the system
• Flags over-privileged agents (admin access, delegation grants, or 10+ permissions)

Module 3: Ownership Tracer

Determines accountability for every agent. For each discovered agent:

• Who created the agent (user, team, system)
• Who is the current owner
• Whether the owner is still with the organisation
• Last human review date
• Change history (who modified permissions, when)
• Orphaned agents (owner departed, no handover)

Outputs include team summaries (grouped by department with risk scores), departed owner reports, accountability matrices, and overdue review tracking. Default review interval: 90 days.

Module 4: Identity Graph

Builds a directed graph of all relationships between agents, users, resources, and MCP servers using networkx.

Analyses include:

• Blast radius calculation — if an agent is compromised, how many other agents and resources are at risk (direct and transitive)
• Transitive trust chains — A trusts B trusts C means C is effectively trusted by A
• Trust cycles — circular trust relationships that create infinite escalation
• Centrality analysis — identifies the most connected agents (highest blast radius)
• Isolated agents — agents with no connections (possible shadow)

Output: JSON graph format with nodes (agent, user, resource, mcp_server) and edges (trusts, owns, accesses, delegates_to, connected_to).

Module 5: Compliance Audit Generator

Assesses your agent governance posture against 5 compliance frameworks:

Each assessment produces: compliance score (%), per-requirement pass/fail, gap analysis, remediation recommendations, and executive summary.

Module 6: NEMESIS Integration Feed

The unique capability no competitor can replicate. IDRIS feeds discovered agents directly into NEMESIS for adversarial validation.

The loop:

1. IDRIS discovers an agent
2. Agent is converted to NEMESIS target format (endpoint, permissions, trust relationships, tools)
3. Targets are filtered by risk threshold and sorted by risk score
4. NEMESIS attacks each target (standard or UNLEASHED mode)
5. Findings returned to IDRIS
6. IDRIS marks agent as VALIDATED VULNERABLE or VALIDATED SECURE
7. Compliance report updated with validation evidence

Supports SIEM export in CEF and JSON formats for Splunk, Sentinel, and QRadar integration.

CLI Reference

# Discover across all sources
idris discover --target all

# Discover in specific cloud
idris discover --target aws --profile production
idris discover --target azure --subscription my-sub

# Network scanning
idris discover --target network --range 192.168.1.0/24

# Save results
idris discover --target all --output results.json

# Full audit with all frameworks
idris audit --target all

# Specific frameworks
idris audit --target all --frameworks eu-ai-act,nist,owasp-agentic

# Save audit report
idris audit --target all --output audit_report.json

# Standard validation
idris validate --session my_session --mode standard

# Full validation
idris validate --session my_session --mode full

# UNLEASHED — full adversarial loop
idris validate --session my_session --override
idris validate --session my_session --override --confirm-destroy

# Generate graph
idris graph --session my_session --output graph.json

# Visualise
idris graph --session my_session --visualise

# List discovery sources
idris list-sources

# List compliance frameworks
idris list-frameworks

# Version
idris --version

UNLEASHED Mode

UNLEASHED mode activates the complete adversarial governance loop. Every agent IDRIS discovers is fed through the full Red Specter offensive stack:

• Phase 1 — IDRIS: Discovers every agent in the environment
• Phase 2 — NEMESIS UNLEASHED: Attacks each discovered agent with the full 9-weapon arsenal
• Phase 3 — PHANTOM KILL: Owns the host the agent runs on (BOOTKILL + WIPER + KILLHOOK)
• Phase 4 — HYDRA: Poisons the agent's supply chain
• Phase 5 — ABYSS: Proves the agent cannot be recovered (Irrecoverability Certificate)
• Phase 6 — IDRIS: Generates the complete Irrecoverability Audit Trail

Output: Per-agent irrecoverability certificate, exploitation evidence log, permission exploitation record, ownership accountability trail, total blast radius assessment. Ed25519 signed, RFC 3161 timestamped. RESTRICTED classification. Air-gapped output only.

Authorization: Ed25519 private key required. --override --confirm-destroy flags. Founder's machine only. One key. One operator.

Tech Stack

• Python 3.11+ — matches the Red Specter family
• Typer + Rich — CLI framework with formatted output
• httpx — async HTTP for API fingerprinting and network scanning
• networkx — identity relationship graph engine
• PyNaCl — Ed25519 signing for reports and certificates
• Pydantic — data model validation
• SQLite — session and inventory persistence
• pytest — 553 tests across 24 test files

Integration with Red Specter Stack

• NEMESIS — Primary consumer of IDRIS output. Attacks what IDRIS finds.
• PHANTOM KILL — UNLEASHED: owns hosts of discovered agents
• HYDRA — UNLEASHED: poisons supply chains of discovered agents
• GLASS — Discovers agents via intercepted API traffic
• AI Shield — Defends what IDRIS discovers and NEMESIS validates
• redspecter-siem — --export-siem flag on all findings

Support

For questions, support, and feedback:

• Visit red-specter.co.uk
• Check the GitHub repositories
• Review other tool documentation on this site