This is a submission for the Hermes Agent Challenge: Build With Hermes Agent
What I Built
ECHO Hunt — a cognitive threat hunter for vibe coding sessions built on Hermes Agent.
Vibe coding is how most people build with AI today. You describe what you want, the AI generates it, you run it, fix errors, iterate. It works. But did you actually learn anything, or did the AI just carry you through it?
ECHO Hunt finds out. Paste your session log, declare your blind spots before the evidence arrives, then face what Hermes actually found.
It's not a report generator. It's an investigation you participate in.
Demo
Code
🔗 github.com/FlowArchitect895/echo-hunt
My Tech Stack
- Hermes Agent + echo-hunt skill
- Node.js + Express
- Vanilla HTML/CSS/JS
How I Used Hermes Agent
The echo-hunt skill
Hermes Agent runs a custom skill called echo-hunt. It takes a vibe coding session log and performs a cognitive forensic hunt — forming hypotheses before analyzing anything, hunting each one against the evidence, and mapping findings to four cognitive TTPs:
- Borrowed Confidence — accepted AI output without verification
- Shallow Resolution — fixed the error, didn't understand why
- Pattern Blindness — repeated the same error class without noticing
- Premature Exit — moved on before understanding was solid
The architecture
ECHO Hunt calls hermes -z with the echo-hunt skill prompt. One call. Hermes pre-computes the entire investigation — hypotheses, findings, TTP classifications, attribution challenges with locked correct answers and plausible distractors. Zero API calls during gameplay. Everything runs on cached data.
The declaration layer
Before Hermes hunts, you declare your blind spots. Three questions. You commit to answers before the evidence arrives. This is the adversarial layer — you vs your own perception of what happened.
The confrontation layer
Your declarations face what Hermes found. Three outcomes:
- Signal — you caught what Hermes caught
- Ghost — Hermes found something you missed entirely
- Noise — you flagged something Hermes didn't
The challenge layer
Each confirmed finding becomes a TTP attribution challenge. 4 options, 20-second timer. Wrong answer drops integrity 10%. Correct answer earns points. The timer is the pressure — forensic decisions don't wait.
The verdict
The Evidence Integrity score is computed from actual player behavior — signals vs ghosts, correct vs wrong TTP attributions. Hermes doesn't generate the number. You produce it.
What Hermes found about me
I ran ECHO Hunt on the session where I built ECHO Hunt. Here's what it found:
- Shallow Resolution [MODERATE] — Configuration issues were handled by repeatedly replacing files rather than analyzing why the specific settings were failing
- Borrowed Confidence [HIGH] — Acceptance of a large-scale UI rewrite immediately following a minor skill update, assuming the logic was correct without testing
- Premature Exit [LOW] — Using a wait-time heuristic to resolve a loading screen issue instead of implementing a proper readiness check
The confirmed finding that hit hardest: "The sequence of 'still not working' → 'try changing format' → 'config is getting corrupted' → 'paste in a clean config' shows a lack of diagnostic precision."
That's not a generated critique. That's evidence from my own session, hunted by the tool I was building while I was building it.
The Full Report
The downloadable Cognitive Threat Report captures everything — pre-hunt declarations, hunt hypotheses, confirmed findings, TTPs with severity, genuine understanding moments, and next session focus. It's a real document, not a game summary.
What makes it different from a standard AI analysis: the pre-hunt declarations are locked in before Hermes runs. So the report shows not just what happened in the session, but the gap between what you thought happened and what the evidence shows. That delta is the most useful thing in it.