Back to search
hermes-dojo
SkillGitHub
Self-improvement system for Hermes Agent. Monitors performance, finds weak skills, fixes them with self-evolution, reports results.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Checked today
Badge for your README
Markdown
[](/listings/yonkoo11--hermes-dojo)HTML
<a href="/listings/yonkoo11--hermes-dojo"><img src="/api/listings/yonkoo11--hermes-dojo/badge.svg" alt="findsafeskills safety rating" /></a>Install
i
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
# Install as a Hermes skill git clone https://github.com/Yonkoo11/hermes-dojo.git cd hermes-dojo ./install.sh # Seed demo data (optional, for testing) cd ~/.hermes/skills/hermes-dojo/scripts python3 seed_demo_data.py --days 7 # Run the full pipeline python3 demo.py --reset
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
SKILL.md · Manifest · 141 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- ---
- name: hermes-dojo
- description: >
- Continuous self-improvement system for Hermes Agent. Analyzes your past sessions
- to find recurring failures and skill gaps, then automatically creates or patches
- skills and runs self-evolution to fix them. Set it to run overnight and wake up
- to a better agent. Use /dojo to start.
- version: 1.0.0
- license: MIT
- metadata:
- author: yonko
- hermes:
- tags: [self-improvement, self-evolution, analytics, meta-agent]
- category: agent-improvement
- requires_toolsets: [terminal]
- allowed-tools: Bash(python3:*) Read Write skill_manage delegate_task session_search memory
- ---
- ## Overview
- Hermes Dojo is your agent's training gym. It reads your past sessions, finds where
- the agent struggles, creates or improves skills to fix those weaknesses, and tracks
- improvement over time.
- The core loop: **measure → identify weakness → fix → evolve → verify → report**
- ## Commands
- - `/dojo` or `/dojo analyze` — Analyze recent sessions for failure patterns
- - `/dojo improve` — Fix the top weaknesses (patch skills + run self-evolution)
- - `/dojo report` — Show current performance metrics and improvement history
- - `/dojo history` — Show learning curve over time
- - `/dojo auto` — Set up overnight cron: analyze + improve + report at 6am
- - `/dojo status` — Quick summary of agent health
- ## Workflow
- ### Step 1: Analyze
- Run `python3 ~/.hermes/skills/hermes-dojo/scripts/monitor.py` to scan recent sessions.
- This reads `~/.hermes/state.db` and produces a JSON report with:
- - Per-tool success/failure rates
- - Error patterns (grouped by tool and error type)
- - User correction signals (messages containing "no,", "wrong", "I meant", "not what I")
- - Skill gap detection (repeated manual tasks with no skill)
- - Session-level metrics (tool calls per session, retry patterns)
- Present the results as a clear dashboard:
- ```
- === Hermes Dojo Analysis ===
- Sessions analyzed: 23 (last 7 days)
- Total tool calls: 156
- Overall success rate: 78%
- Top Weaknesses:
- 1. terminal_run: 73% success (12 failures) — common error: "command not found"
- 2. web_extract: 81% success (4 failures) — common error: "timeout"
- 3. No skill for: CSV parsing (requested 4 times)
- User Corrections Detected: 7
- - 3x wrong file path
- - 2x wrong command syntax
- - 2x misunderstood request
- ```
- ### Step 2: Improve
- For each identified weakness, decide the fix:
- **A) Existing skill fails → patch it:**
- 1. Read the current skill's SKILL.md
- 2. Analyze the failure patterns from Step 1
- 3. Use `skill_manage` with action "patch" to add error handling, better instructions, or edge case coverage
- 4. Log the change
- **B) No skill exists for a recurring need → create one:**
- 1. Analyze the session patterns where this capability was needed
- 2. Use `skill_manage` with action "create" to make a new skill
- 3. Include specific instructions based on what worked in past sessions
- 4. Log the creation
- **C) Skill exists but needs deeper improvement → run self-evolution:**
- 1. Run: `cd ~/.hermes/hermes-agent-self-evolution && .venv/bin/python3 -m evolution.skills.evolve_skill --skill <name> --hermes-repo ~/.hermes --iterations 5 --eval-source synthetic`
- 2. This uses GEPA to analyze execution traces and propose targeted improvements
- 3. Review the evolution output — accept if score improved
- 4. Log before/after scores
- ### Step 3: Verify
- After improvements:
- 1. Re-run `monitor.py` to compute new metrics
- 2. Compare before/after success rates
- 3. If improvement < 5%, flag for manual review
- 4. Store results in metrics history
- ### Step 4: Report
- Run `python3 ~/.hermes/skills/hermes-dojo/scripts/reporter.py` to generate a report.
- Format for Telegram delivery:
- ```
- 🥋 Hermes Dojo — Overnight Report
- 📊 Analyzed: 23 sessions, 156 tool calls
- 📈 Overall: 78% → 85% (+7%)
- ✅ Improved:
- • terminal_run: 73% → 96% (added PATH check)
- • web_extract: 81% → 95% (added retry + timeout)
- 🆕 New skill created:
- • csv-handler: for CSV parsing you keep asking about
- ⚠️ Still working on:
- • File path resolution (need more examples)
- 📉 Learning curve: 71% → 78% → 85% (3 days)
- ```
- ### Step 5: Track
- Run `python3 ~/.hermes/skills/hermes-dojo/scripts/tracker.py save` to persist metrics.
- Run `python3 ~/.hermes/skills/hermes-dojo/scripts/tracker.py history` to show the learning curve.
- ## Setting Up Overnight Cron
- When the user says `/dojo auto`:
- Set up a cron job that runs at 6:00 AM daily:
- "Run /dojo analyze, then /dojo improve on the top 3 weaknesses, then send the report to my home channel."
- This makes the agent literally improve while you sleep.
- ## Important Notes
- - Never modify bundled skills (in hermes-agent/skills/). Only modify user skills in ~/.hermes/skills/
- - Self-evolution requires ~/.hermes/hermes-agent-self-evolution to be cloned
- - Metrics are stored in ~/.hermes/skills/hermes-dojo/data/metrics.json
- - Each analysis run is timestamped so you can track improvement over days/weeks
README.md · README · 127 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
- # Hermes Dojo
- **Your agent, getting better while you sleep.**
- Hermes Dojo is a self-improvement system for [Hermes Agent](https://github.com/NousResearch/hermes-agent) that watches how your agent performs, finds its weakest skills, fixes them with self-evolution, and shows you the results.
- ## The Problem
- Your AI agent makes the same mistakes every day. You correct it, it forgets next session. Skills exist but nobody knows which ones work well and which ones silently fail. Self-evolution exists but nobody uses it because there's no signal about WHAT to evolve.
- ## The Solution
- Dojo closes the feedback loop:
- ```
- measure → identify weakness → evolve → measure again → report
- ```
- It turns "the agent that grows with you" from a tagline into reality.
- ## How It Works
- ### 1. Performance Monitor
- Reads your agent's session logs from `state.db`. Identifies failures: tool errors, retry loops, user corrections ("no, I meant..."), explicit complaints. Tracks per-skill success rates.
- ### 2. Weakness Analyzer
- Categorizes root causes and ranks improvement opportunities:
- - "web_extract fails 100% of the time. Root cause: no retry logic for timeouts"
- - "No skill exists for CSV parsing, but user asked for it 5 times"
- ### 3. Auto-Fixer
- For each weakness:
- - Skill exists but fails → patch it with targeted error handling
- - No skill exists → create one based on session patterns
- - Run self-evolution (GEPA) on weak skills for measurable improvement
- ### 4. Reports
- Generates reports for CLI or Telegram delivery with deltas, sparklines, and actionable summaries.
- ### 5. Learning Curve
- Stores daily metrics. Shows improvement over days/weeks. Proof that the agent is actually growing.
- ## Quick Start
- ```bash
- # Install as a Hermes skill
- git clone https://github.com/Yonkoo11/hermes-dojo.git
- cd hermes-dojo
- ./install.sh
- # Seed demo data (optional, for testing)
- cd ~/.hermes/skills/hermes-dojo/scripts
- python3 seed_demo_data.py --days 7
- # Run the full pipeline
- python3 demo.py --reset
- ```
- ## Commands (via Hermes Agent)
- | Command | What it does |
- |---------|-------------|
- | `/dojo analyze` | Analyze recent sessions for failures |
- | `/dojo improve` | Fix weakest skills + run self-evolution |
- | `/dojo report` | Generate improvement report |
- | `/dojo history` | Show learning curve over time |
- | `/dojo auto` | Set up overnight cron (analyze + improve + report) |
- ## Architecture
- ```
- hermes-dojo/
- ├── SKILL.md # Main orchestrator (Hermes skill format)
- ├── scripts/
- │ ├── monitor.py # Reads session SQLite, computes metrics
- │ ├── analyzer.py # Categorizes failures, ranks weaknesses
- │ ├── fixer.py # Patches skills, creates new ones, runs evolution
- │ ├── reporter.py # Generates CLI/Telegram reports
- │ ├── tracker.py # Stores/retrieves learning curve data
- │ ├── seed_demo_data.py # Demo data generator
- │ └── demo.py # Full pipeline demo runner
- ├── references/
- │ └── failure_patterns.md
- └── data/
- └── metrics.json # Historical performance data
- ```
- ## Hermes Features Used
- | Feature | How Dojo Uses It |
- |---------|-----------------|
- | Skills system | Dojo IS a skill; it creates/patches other skills |
- | Self-evolution (GEPA) | Evolves weak skills via DSPy optimization |
- | Session search | Reads past sessions to identify failure patterns |
- | Cron scheduler | Runs overnight improvement cycle |
- | Multi-platform | Morning report on Telegram |
- | skill_manage | Creates and patches skills programmatically |
- ## Requirements
- - [Hermes Agent](https://github.com/NousResearch/hermes-agent) v0.2.0+
- - Python 3.10+
- - [hermes-agent-self-evolution](https://github.com/NousResearch/hermes-agent-self-evolution) (for GEPA)
- ## Built for
- Nous Research Hermes Agent Hackathon (March 2026)
- ## Contributing
- Issues and pull requests welcome. The analyzer in particular benefits from
- real-world failure patterns — if you see a false-positive recommendation or a
- weak-skill misclassification, please open an issue with the offending tool
- name and top error string.
- Thanks to external contributors:
- - [@eggressive](https://github.com/eggressive) — root-cause classification and
- description-based gap detection in the weakness analyzer (#3)
- ## License
- MIT — see [LICENSE](LICENSE).
- ---
- Built on [Hermes Agent](https://github.com/NousResearch/hermes-agent) by [Nous Research](https://nousresearch.com)
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.