Skip to content

Hermes Agent Integration

SecOpsAI supports Hermes Agent as a local-first telemetry source alongside OpenClaw and host adapters for macOS, Linux, and Windows.

The Hermes adapter is read-only. It does not execute Hermes, run tools, install packages, call model APIs, or modify Hermes configuration. It reads local telemetry that Hermes already writes and converts it into SecOpsAI's shared event schema for detection, triage, and correlation.

What SecOpsAI Collects

By default, secopsai refresh --platform hermes reads:

Source Path Purpose
Hermes command history ~/.hermes/.hermes_history Operator prompts and command-like history entries
Agent logs ~/.hermes/logs/agent.log Agent and tool execution status
Error logs ~/.hermes/logs/errors.log Agent errors and failed tool calls
Gateway logs ~/.hermes/logs/gateway.log Gateway/channel/session lifecycle events
Session state ~/.hermes/sessions/sessions.json Session metadata, platform, and chat type
Request dumps ~/.hermes/sessions/request_dump_*.json Redacted API failure metadata
Gateway state ~/.hermes/gateway_state.json Local gateway state

SecOpsAI intentionally does not read ~/.hermes/auth.json, ~/.hermes/.env, or provider credential stores.

Quick Start

cd ~/secopsai
source .venv/bin/activate

secopsai refresh --platform hermes
secopsai list --platform hermes
secopsai show SCX-XXXX

To combine Hermes with OpenClaw and host telemetry:

secopsai refresh --platform hermes,openclaw,macos
secopsai correlate

If Hermes is installed somewhere other than ~/.hermes, set:

HERMES_HOME=/path/to/hermes-home secopsai refresh --platform hermes

Detection Coverage

The first Hermes rules are intentionally high-signal:

Rule Detects
RULE-120 Hermes Dangerous Tool Call Pipe-to-shell, reverse-shell, destructive, or staged interpreter command patterns
RULE-121 Hermes Credential Exfiltration Credential-file discovery plus archive/upload, GitHub token handling, or outbound transfer behavior
RULE-122 Hermes Request Dump Secret Leak Token-like material that remains visible after SecOpsAI redaction

Routine model errors, missing credits, gateway startup messages, and normal Telegram/channel activity are collected for context but are not findings by default.

Redaction And Safety

The Hermes adapter redacts token-like values before normalized events are stored. It also summarizes request dumps instead of preserving raw message bodies or headers. Findings include safe command/evidence snippets, not raw credentials.

Keep these defaults:

  • Do not sync raw Hermes logs to third-party systems without review.
  • Use SecOpsAI findings and redacted evidence for triage.
  • Rotate credentials if RULE-121 or RULE-122 fires.
  • Treat channel/session identity as part of incident context.

Operator Workflow

  1. Run secopsai refresh --platform hermes.
  2. Review Hermes findings with secopsai list --platform hermes.
  3. Inspect evidence with secopsai show <finding-id>.
  4. Correlate with OpenClaw or host telemetry if the same user, session, path, or time window appears elsewhere.
  5. If credential access is suspected, rotate affected tokens and temporarily disable the relevant Hermes channel/toolset while investigating.

Troubleshooting

If Hermes is not detected:

ls ~/.hermes
HERMES_HOME=~/.hermes secopsai refresh --platform hermes

If no findings appear, that can be healthy. SecOpsAI still collected context, but only high-confidence Hermes abuse patterns are promoted to findings.