All plugins

remedigenius

Read-only AWS security triage desk on Hermes: prioritizes what Security Hub, GuardDuty and Inspector already found, verifies with read-only calls, and drafts IaC fixes a human approves. Disclosure — runs the rgtriage engine locally (subprocess or local socket), which reads AWS with your configured read-only credentials and keeps state in a per-profile home under your Hermes home (existing ~/.rgtriage installs keep it); Terraform validation needs your approval and installs only providers already in the module's lock file or an operator-vetted plugin dir; findings and chat go to the model provider you configure; local execution traces are on by default (RGTRIAGE_TRACES=off disables); no telemetry, no self-update.

RajaRengaToolsv0.9.3Updated 1 day ago

About

Read-only AWS security triage desk on Hermes: prioritizes what Security Hub, GuardDuty and Inspector already found, verifies with read-only calls, and drafts IaC fixes a human approves. Disclosure — runs the rgtriage engine locally (subprocess or local socket), which reads AWS with your configured read-only credentials and keeps state in a per-profile home under your Hermes home (existing ~/.rgtriage installs keep it); Terraform validation needs your approval and installs only providers already in the module's lock file or an operator-vetted plugin dir; findings and chat go to the model provider you configure; local execution traces are on by default (RGTRIAGE_TRACES=off disables); no telemetry, no self-update.

rg-triage

The Hermes plugin half of RemediGenius Community Edition:
a read-only AWS security triage desk. It gives the agent one structured tool over the rgtriage
engine, shows the attention board at session start, and adds /triage and /rate.

This plugin is one piece of the desk. The full installation — profile, skills, dashboard tabs and
the engine — is described in the repository README.
Installing this plugin alone adds the tools to an existing Hermes; it expects the
remedigenius-community package (pinned in plugin.yaml) to be importable.

What it registers

  • Tools: rg_triage (scan, list, show, card, rate, doctor, onboarding, knowledge — all read-only)
    and rg_analytics (read the desk's own redacted traces).
  • Slash commands: /triage, /rate.
  • Hooks: board summary before the first model call, redaction of traces, answer formatting.

What this changes in your Hermes profile

Wherever this plugin is enabled, it changes how the agent behaves:

  • Approvals. decide, remember, github publish and validate go through Hermes's approval
    prompt before they run. The validate prompt names the module and the providers terraform init
    may install (only ones already in the module's .terraform.lock.hcl).
  • Write guard. write_file and patch are blocked everywhere except the desk's proposals folder.
    The agent drafts changes; it never edits your repositories directly.
  • Streaming. The plugin holds streaming for every reply (hold_stream): critical reports are
    rendered from desk evidence and redacted before you see them, not streamed token by token.
  • Local traces, on by default. Every turn (messages and tool results, redacted) is written to a
    local trace under rg-traces in the state home (see below), kept 30 days, read by rgtriage improve
    and the analytics tab. Nothing is sent anywhere. Set RGTRIAGE_TRACES=off to disable it entirely.
  • Read guard. Credential stores (~/.aws, ~/.ssh, keys, .env, Terraform state) and whole-home
    or drive-root searches are blocked.

What it does on your machine

  • Runs python -m rgtriage … as a subprocess, or talks to a running desk over a local socket.
    The engine reads AWS through the credentials you configure (a read-only role is the documented
    setup) and never writes to AWS. Every fix is a proposal a person reviews.
  • Reads and writes its own state in a per-profile home (rgtriage under your Hermes home), so two
    Hermes profiles never share one desk. Installs that already have state under ~/.rgtriage keep
    using it until you move it; RGTRIAGE_HOME or the installer's local.json override both. A
    proposals folder for drafted IaC changes lives beside the state.
  • Validation (validate, and the check before github publish) runs terraform init/validate/fmt,
    tflint and cfn-lint on a scratch copy — never plan or apply. Model-written proposals cannot add or
    change terraform {} / required_providers blocks or .terraform.lock.hcl; init installs only
    providers already in the module's lock file, or from an operator-vetted RGTRIAGE_TF_PLUGIN_DIR
    (no registry access in that mode). The tools run with an allowlisted environment, so model and
    cloud keys never reach them or the providers they load.
  • Optional environment overrides: RGTRIAGE_PYTHON, RGTRIAGE_SRC, RGTRIAGE_HOME,
    RGTRIAGE_SOCKET, RGTRIAGE_PROPOSALS, RGTRIAGE_STALE_HOURS, RGTRIAGE_TRACES,
    RGTRIAGE_TRACES_DIR, RGTRIAGE_TF_PLUGIN_DIR. None are required.
  • The plugin holds no AWS or model credentials. The model key belongs to your Hermes profile;
    findings and chat go to the model provider you configure.
  • No telemetry, no network calls of its own, no self-update.

Limits

  • No detectors of its own: it works with what Security Hub, GuardDuty, Inspector and your other
    AWS services already report.
  • AWS only, and early software: see docs/LIMITS.md.

Apache-2.0.

README from the reviewed commit b35dac3; it updates when the author re-pins.

More by RajaRenga

3 plugins by RajaRenga in the catalog.

RajaRengaiacg-boardWeb dashboard tab showing the IaC Genius mission board: findings, validation evidence with tool + version, repair attempts spent/allowed, disposition and next owner. It renders the snapshot the iacg-desk service writes and never invents live state; an empty state is shown until a snapshot exists. Disclosure — read-only dashboard tab: reads the iacg-desk board snapshot (board.json/board.html) over the local Unix socket or the shared data dir, renders it in a sandboxed iframe (scripts allowed, no same-origin access) and accepts messages only from that iframe; registers no agent tools or hooks, makes no network calls, runs no subprocess, no telemetry, no self-update. Disclosure — talks to the local iacg desk service over a Unix socket (or runs python -m iacg_desk as a subprocess); the desk's terraform init downloads providers read-only from the public Terraform registry into a shared cache; while the plugin is enabled it refuses terraform/tofu apply and cloud write commands in every terminal call of every session, not only IaC Genius missions; approve, confirm and deliver each raise Hermes' approval prompt so a person must grant them; model-supplied tree/out paths are confined to the desk work dir and validation tools run with a secret-free environment; delivery pushes a branch and opens a pull request on GitHub with the IACG_GITHUB_TOKEN secret only after the gates pass and a human approves; no telemetry, no self-update.CommunityDesktop appRajaRengaiacg-deskIaCGenius runs deterministic Terraform/OpenTofu missions on Hermes: a decision ledger asks the costly or irreversible questions before spend, freezes the agreed scope as one build-sheet hash, grounds every argument on the pinned binary provider schema, runs terraform validate / tflint / checkov / secret scan as evidence, repairs within a bounded budget, and delivers a reviewable PR whose body is the ledger. The model proposes; the toolchain decides. Disclosure — talks to a local Unix socket served by the iacg_desk desk service (or runs python -m iacg_desk as a subprocess); reads its own local.json + IACG_* config paths (not secrets); the desk's terraform init fetches providers read-only from the public registry into a shared cache; hard-refuses terraform/tofu apply and cloud write commands at pre_tool_call, and escalates approve/confirm/deliver to Hermes's human approval gate so a person must grant them (never model-self-approved); model-supplied tree/out paths are confined to the desk work dir; validation tools run under a secret-free allowlisted env; holds no cloud credentials and never changes cloud state; GitHub delivery uses a token read through Hermes's secret scope only after the gates and a human approval, then reads the commit back; the board briefing is injected via pre_llm_call on the first turn; no telemetry, no self-update.Communitytools

Developer Mode

Explore built-in Hermes toolsets and actions.

HermesAgent

Developer?

Create a Hermes extension for Desktop or agent workflows.

Developer Section