Persona System¶

Siyarix leverages a Persona System to dynamically adapt the LLM's response style, focus areas, and depth of expertise. Personas act as dynamic prompt preambles injected into the system prompt before each AI interaction.

Personas are defined in external files: - Built-in personas: src/siyarix/data/personas/ (shipped with package, read-only) - Custom personas: ~/.siyarix/data/personas/custom/ (user-created, writable) - Load order: User directory checked first, falls back to built-in

The system supports 10 specialised security-domain personas plus 3 special modes.

Quick Reference¶

Command	Effect
`/persona`	View currently active persona
`/persona list`	List all available personas
`/persona <name>`	Switch to a named persona
`/persona auto`	Smart auto-selection based on request
`/persona universal`	Full-spectrum cybersecurity professional
`/persona none`	Neutral mode (no persona framing)

Available Personas¶

Named Personas¶

Name	Label	Focus Area	Key Tooling & Frameworks
`redteam`	Red Team / Offensive Security	Adversary emulation, exploitation, evasion	PTES, Cobalt Strike, BloodHound, Chisel
`blueteam`	Blue Team / Defensive Security	Detection engineering, threat hunting, IR	Sigma, YARA, Velociraptor, osquery, Zeek
`purpleteam`	Purple Team / Collaborative Security	Attack validation, detection coverage	Atomic Red Team, Caldera, ATT&CK
`dfir`	DFIR / Digital Forensics	Forensic analysis, malware triage	Volatility, Plaso, Autopsy
`threatintel`	Threat Intelligence (CTI)	Threat research, IoC extraction	Diamond Model, STIX, MISP, OpenCTI
`cloudsec`	Cloud Security (CloudSec)	AWS/Azure/GCP hardening, containers	Prowler, ScoutSuite, trivy
`appsec`	Application Security (AppSec)	Secure code review, SAST/DAST	OWASP Top 10, Semgrep, CodeQL
`networksec`	Network Security (NetSec)	Network architecture, firewalls	Zeek, Suricata, Wireshark
`governance`	Governance / GRC	Compliance, risk management, audits	ISO 27001, SOC 2, NIST CSF
`securityexplorer`	Security Explorer / Research	Vuln research, reverse engineering, CTFs	Ghidra, AFL++, CVE research

Special Modes¶

Name	Behaviour
`auto`	Smart Select: Analyses your prompt and uses LLM reasoning to pick the best-fit persona
`universal`	The Swiss Army Knife: Combines all domain expertise into one balanced professional
`none`	Neutral Mode: Strips all persona framing, uses bare system prompt

Data File Layout¶

Personas are defined across two file types:

Persona Index (`index.json`)¶

{
  "red team": {
    "name": "red team",
    "label": "Red Team / Offensive Security",
    "description": "Adversary emulation, penetration testing, exploitation, C2 operations, evasion",
    "file": "red-team.md"
  }
}

Persona Prompt Files (`red-team.md`, `blue-team.md`, etc.)¶

Each file contains the persona's expertise paragraph, loaded at runtime:

You are an elite red-team operator who conducts realistic adversary emulation...

Custom Personas (`~/.siyarix/data/personas/custom/*.json`)¶

Users can add custom personas as JSON files with name, label, description, and prompt fields:

{
  "name": "my-specialist",
  "label": "My Specialist Persona",
  "description": "Custom persona for specialised tasks",
  "prompt": "You are a specialist in..."
}

How It Works Under the Hood¶

Prompt Construction¶

The build_persona_prompt() function in personas.py loads persona data from external files and generates the preamble:

def build_persona_prompt(persona_name: str) -> str:
    p = get_persona(persona_name)
    if not p or persona_name == "none":
        return ""

    if persona_name == "auto":
        lines = ["## Active Persona: Auto (Smart Select)"]
        lines.append("Analyse the user's request below and automatically adopt...")
        for name, pp in _get_personas().items():
            if name not in ("auto", "none"):
                lines.append(f"  - **{pp['label']}**: {pp['description']}")
        return "\n".join(lines)

    return f"## Active Persona: {p['label']}\n{p['prompt']}"

Logic Summary¶

Active Setting	Initial Request	Follow-up (Compact)
`none`	`system-neutral.md` only	`compact-neutral.md`
Named (e.g., `redteam`)	Persona preamble + `system.md` + `RULES.md`	Persona label + `compact.md`
`universal`	Universal preamble + `system.md` + `RULES.md`	Persona label + `compact.md`
`auto`	Auto preamble (lists options) + `system.md` + `RULES.md`	Persona label + `compact.md`

Data Loading¶

Personas are loaded by personas.py through data_loader.py:

Load index.json from data/personas/ (user dir first, built-in fallback)
Load each .md prompt file referenced in the index
Load custom personas from ~/.siyarix/data/personas/custom/*.json
Merge all into a single cache

The auto and none special modes are always defined in code (not external files) since they have special logic.

Persona Data Model¶

PERSONAS = {
    "red team": {
        "name": "red team",
        "label": "Red Team / Offensive Security",
        "description": "Adversary emulation...",
        "prompt": "You are an elite red-team operator...",
    },
    # ... built-in personas loaded from data files
    "my-custom": {
        "name": "my-custom",
        "label": "My Custom Persona",
        "description": "...",
        "prompt": "...",
    },  # Loaded from ~/.siyarix/data/personas/custom/
    "auto": { ... },    # Special mode (always in code)
    "none": { ... },    # Special mode (always in code)
}

Key Functions¶

get_persona(name: str) -> dict | None             # Case-insensitive lookup
get_all_personas() -> dict                        # Full dict (including special modes)
list_personas() -> list[dict]                     # Standard named personas only
list_all_personas() -> list[dict]                 # All personas (including special + custom)
build_persona_prompt(name: str) -> str            # Generate the prompt preamble text
reload_personas() -> None                         # Force reload from external files

Persona lookups are fuzzy! _normalize_persona_key() strips spaces, hyphens, and underscores. /persona red-team works the same as /persona redteam.

Configuration¶

Set a persistent default in settings.toml:

persona = "redteam"
additional_system_message = "Always output findings as a Markdown table"

Custom instructions are appended after the system prompt and rules document.

Integration with Interaction Modes¶

Personas apply to all LLM-dependent modes (integrated, autonomous, hybrid). In registry or offline modes, the persona system is bypassed entirely (no LLM calls).

Module / Function	Path	Purpose
`_load_personas()`	`src/siyarix/personas.py`	Loads all personas from external files + custom dir
`build_persona_prompt()`	`src/siyarix/personas.py`	Assembles the persona preamble text
`list_custom_personas()`	`src/siyarix/data_loader.py`	Lists custom personas from user directory
`load_text()` / `load_json()`	`src/siyarix/data_loader.py`	Generic data loaders with fallback
`system.md`	`src/siyarix/data/prompts/system.md`	Full system prompt template
`RULES.md`	`src/siyarix/data/rules/RULES.md`	All operational rules