Analyze recent sessions for performance metrics.

Load historical metrics.

Generate prioritized improvement recommendations.

Generate a formatted improvement report.

Try to find which skill is responsible for a given tool.

Pre-load the description line from each SKILL.md for fuzzy matching. Uses manual parsing so that there is no dependency on PyYAML.

Higher score = more urgent to fix.

Suggest a fix based on the error type.

Apply fixes from the plan. Returns list of applied improvements.

Classify an error into a fix strategy category.

Generate a complete fix plan from analyzer recommendations.

Print the learning curve.

Save a metrics snapshot.

Seed the database with realistic demo data. Args: days: Number of days of data to seed clear: Clear previously seeded data first

Build skill content — use specific template if available, otherwise generate a targeted skill based on the error type and failure data.

Classify the root cause of an error. Returns whether it's fixable by a skill.

CLI-formatted report.

Load OPENROUTER_API_KEY from Hermes .env file or environment.

Telegram-formatted report (markdown).

Convert a tool name to a valid skill name.

Classify a tool result as success or failure, return (is_error, error_type).

Detect when the same tool is called multiple times in quick succession (retry loop).

Scan all installed skills and return name -> path mapping.

Generate a new skill creation with specific, actionable instructions. Unlike boilerplate templates, this builds skills with real bash commands,

Generate a skill patch instruction for a recommendation.

Print a human-readable dashboard.

Print the fix plan in human-readable format.

Print recommendations in human-readable format.

Run the full Dojo demo pipeline.

Run self-evolution on a skill via the hermes-agent-self-evolution CLI.

Check whether any installed skill already covers a capability. Searches skill names and their SKILL.md descriptions for keyword matches.

Pre-populate metrics history with a realistic multi-day improvement arc. Shows what happens when Dojo runs nightly: Day 1-2: baseline (bad),