Fjord Audit

What Fjord Audit does

The audit measures skills (count, category and files over 20 KB), memory files MEMORY.md and USER.md against soft display budgets, plugin directories, profiles, a config snapshot with the model, provider, max_turns and compression keys, and gateway pid, lock and state files. The README calls it filesystem truth only, and says it will not claim that a profile is online, so you should pair it with hermes doctor for connectivity.

It runs three ways: CLI commands (hermes fjord scan, score, version and selftest, with a --home option and --json output), agent tools named fjord_scan and fjord_score, and slash commands such as /fjord scan. Tool handlers always return JSON and report failures as {ok: false, error, version} instead of raising into the agent loop. PyYAML is optional and adds richer config.yaml snapshots. The README gives the version as 1.1.0.

Key features

• Skill count, category and size checks
• Memory file checks against soft budgets
• Plugin, profile, config and gateway-state snapshots
• Graded score with recommendations
• CLI, agent tools and slash commands
• No network and no LLM

When to use it

• Check whether a Hermes profile has too many skills or oversized memory files
• Compare the structure of several profiles with --home
• Produce JSON audit output for scripts

Who it is for: Hermes Agent users with several profiles or a large skills and memory footprint who want a quick structural health check.

How it fits with Hermes Agent

Built for Hermes Agent as a plugin that adds a hermes fjord command, two agent tools and a /fjord slash command.

How to install Fjord Audit

These commands are copied from the project's README. Check the repository for the latest steps before you run them.

hermes plugins install smfworks/hermes-plugin-fjord-audit
hermes plugins enable fjord-audit
pip install 'hermes-plugin-fjord-audit[yaml]'

Requirements: PyYAML is optional, and without it the scan still succeeds but sets config.parse_error

FAQ