decision-brief

Decision Brief — a human decision layer for Hermes Agent

Decision Brief helps Hermes pause at consequential choices and present a compact, understandable decision before continuing. It applies to both engineering/agent decisions and creative workflows such as image generation.

What it does

• asks only when a decision is consequential (ambiguous, hard to reverse, or user-preference-dependent)
• explains the decision compactly: Decision / What I found / Why it matters / Recommendation
• provides a recommendation when evidence supports one
• uses Hermes's native clarify panel — it never invents its own UI
• supports creative decisions around image generation (lighting, composition, mood) before image_generate
• preserves native clarify choices, order, and multi_select — always
• never captures hidden reasoning
• never stores raw user content (prompts, tool results, stream text)

Example 1 — Agent decision

User: "Set up auth — keep Supabase or custom?"

Decision Brief:
Decision: Authentication architecture
What I found: The choice affects implementation complexity,
maintenance, and integration effort.
Why it matters: Changing auth architecture later can be expensive.
Recommendation: Keep Supabase if the existing stack already depends on it.

[Keep Supabase] [Build custom]

Example 2 — Creative decision

User: "Create a cinematic full-body photograph outside a luxury hotel."

Decision Brief can surface a meaningful creative choice such as:

Night vs. golden hour?

Why it matters: The lighting changes the atmosphere,
contrast, and perceived luxury of the scene.

[Golden hour — recommended]
[Night]

The creative decision is skill-guided. Provider-dependent live generation was unavailable during validation because the provider returned HTTP 402.

Validation status

13 tests — PASS
python -m compileall -q . — PASS
hermes plugins doctor — PASS
hermes plugins validate — PASS

Live provider-dependent clarify/image-generation validation was blocked by OpenRouter HTTP 402 during the validation window. The plugin does not claim those provider flows were live-tested.

Architecture

User request
    ↓
Hermes skill guidance (bundled decision-brief skill)
    ↓
Consequential decision detected
    ↓
Native Hermes clarify
    ↓
User approval
    ↓
Hermes continues normally

Plugin hook:

pre_tool_call
    ├── clarify → lightweight question enrichment (choices untouched)
    └── image_generate → observed integration point (never blocked or rewritten)

Image generation is not automatically blocked or rewritten by the plugin. The skill is responsible for getting the user's approved creative direction into the image_generate prompt.

Install

Portable / catalog install:

hermes plugins install <owner>/decision-brief
hermes plugins enable decision-brief

Local development:

cp -r decision-brief <HERMES_HOME>/plugins/decision-brief
hermes plugins enable decision-brief

Then, interactively:

/decision-brief:decision-brief
<ask a consequential question — the agent will ask one focused clarify, or decide solo>

Notes

• /decision-brief-probe shows content-free diagnostics (event counts, hashed scope/turn tokens, current-vs-other session split). Profile-scoped — the profile's gateway is shared across its sessions.
• No telemetry, no external services, no persistent conversation logs, no hidden reasoning capture.

README from the reviewed commit 1722b6d; it updates when the author re-pins.