Hermes Plugin for Omni

[wysie]

Put together a plugin for Hermes! https://github.com/Wysie/hermes-omni-plugin

[fajarhide]

Put together a plugin for Hermes! https://github.com/Wysie/hermes-omni-plugin

Really appreciate you putting this together, this is honestly one of the more thoughtful integrations I’ve seen so far.

You’re not just wrapping OMNI, you actually aligned with the native Hermes execution model. Using transform_terminal_output instead of porting omni_cmd directly from OpenClaw is the right call, and it shows you understand the architectural differences between the two ecosystems.

Also, _enhance_stats_output() is a genuinely interesting addition. Translating OMNI stats into something meaningful for subscription-backed providers (like Codex / ChatGPT), especially the idea of context_hygiene_not_direct_bill_reduction is a subtle but important distinction. That’s original thinking, not just glue code.

The safety layer also stands out. Expanding to 22 dangerous env vars vs the usual ~13 baseline is a solid improvement.

That said, there are two things we should fix before merging:

1. Fragile stats parsing
   _enhance_stats_output() currently relies on regex parsing, which is going to break sooner or later. Since omni stats --json is already available (since v0.5.6-rc1), we should switch to that as the source of truth. It’ll make the transformation deterministic and future-proof.

2. Missing hash validation in _tool_rewind()
   Right now the hash input is passed straight into the shell without validation. We need to enforce a strict format check (length + charset) before execution, otherwise this opens the door for argument injection.

3. Fix _parse_kb() binary vs decimal inconsistency (use 1024)
   OMNI uses binary (1024), but this function mixes decimal conversion → causing consistent drift (~2–3%). Aligning with OMNI’s binary (1024) convention ensures consistent and correct calculations.

4. Update Python requirement to >=3.10
   pyproject.toml allows >=3.8, but code uses dict | None (requires 3.10+). The codebase uses modern type hints (e.g., dict | None) that are not supported in Python <3.10. Updating the requirement avoids runtime errors and confusion for users.

5. Add config backup for /omni reset-config
   The reset command currently overwrites configuration without recovery. Creating a backup before reset prevents accidental data loss and improves safety.

[wysie]

Thanks again for the detailed suggestions. They were very helpful.

I've addressed the maintainer feedback in the Hermes OMNI plugin and pushed the changes here:
wysie/hermes-omni-plugin@6f95dcc

Changes made:

• Replaced fragile enhanced-stats parsing with omni stats --json as the source of truth, with the old human-readable parser kept only as a fallback for older OMNI versions.
• Added rewind hash validation before invoking omni rewind.
• Updated _parse_kb() and fallback estimates to use binary units (1024) instead of decimal (1000).
• Confirmed Python requirement is already >=3.10.
• Added timestamped config backups before /omni reset-config overwrites the config.
• Updated README docs for JSON stats, rewind validation, and config backups.
• Added regression tests for the above.

Verification:

• Local tests: 20 passed
• Build: passed
• Fresh install from GitHub into a clean venv: passed
• GitHub Actions CI: passed

CI run:
https://github.com/wysie/hermes-omni-plugin/actions/runs/25269837429

Appreciate the review. This made the plugin much more robust.

[fajarhide]

Thanks again for the detailed suggestions. They were very helpful.

I've addressed the maintainer feedback in the Hermes OMNI plugin and pushed the changes here: wysie/hermes-omni-plugin@6f95dcc

Changes made:

• Replaced fragile enhanced-stats parsing with omni stats --json as the source of truth, with the old human-readable parser kept only as a fallback for older OMNI versions.
• Added rewind hash validation before invoking omni rewind.
• Updated _parse_kb() and fallback estimates to use binary units (1024) instead of decimal (1000).
• Confirmed Python requirement is already >=3.10.
• Added timestamped config backups before /omni reset-config overwrites the config.
• Updated README docs for JSON stats, rewind validation, and config backups.
• Added regression tests for the above.

Verification:

• Local tests: 20 passed
• Build: passed
• Fresh install from GitHub into a clean venv: passed
• GitHub Actions CI: passed

CI run: https://github.com/wysie/hermes-omni-plugin/actions/runs/25269837429

Appreciate the review. This made the plugin much more robust.

Well done. This is perfect! Thanks for the incredibly fast turnaround and for addressing all the points. The plugin looks solid and production-ready now. Great job! Thank you for helping build the OMNI ecosystem at Hermes Agent.

[wysie]

Thanks! Happy to contribute as I'm also using Omni (and Hermes) extensively.

[fajarhide]

Thanks! Happy to contribute as I'm also using Omni (and Hermes) extensively.

Thanks for use OMNI.

How effective is your performance environment after using Omni? Have you tried using other tools besides Omni and how do they compare?

[wysie]

Submitted a PR to updates the readme #75

[fajarhide]

Submitted a PR to updates the readme #75

Nice, thanks to contributed