hermes - 💡(How to fix) Fix [Bug]: Hermes unexpectedly modifies its own skills/system prompts during normal task execution, even when the user request is unrelated to agent configuration.

Bug Description

Hermes unexpectedly modifies its own skills/system prompts during normal task execution, even when the user request is unrelated to agent configuration.

This causes the agent to rewrite, edit, or optimize its own skill definitions instead of only modifying project/business code.

The behavior is problematic because skills should generally be treated as immutable system assets unless the user explicitly requests skill modification.

Observed impact:

• Existing skill prompts are unexpectedly changed
• Agent behavior becomes inconsistent across sessions
• Skills may drift over time and break established workflows
• Prompt engineering / system rules become unstable

Expected behavior: Hermes should treat skill files (e.g. .hermes/skills/**) as read-only by default and avoid modifying them unless the user explicitly asks to update a skill.

No error message is shown — the issue is behavioral.

Steps to Reproduce

1. Create or configure a reusable Hermes skill (e.g. a Vue/UI development skill).
2. Run hermes chat
3. Ask Hermes to implement a feature or modify project code unrelated to skill configuration.
4. Observe that Hermes edits files inside .hermes/skills/ or modifies its own prompt/skill definitions without explicit instruction.

Example:

• User request: "Build a form component using the existing UI framework"
• Unexpected behavior: Hermes modifies the underlying skill prompt itself instead of only generating or updating project code.

Expected Behavior

Hermes should not modify its own skills, prompts, or system configuration during normal execution.

Skill files should be treated as immutable/read-only unless:

• the user explicitly asks to modify a skill
• the task directly targets agent configuration

By default, Hermes should prioritize changing project/business code instead of self-modifying behavior.

Actual Behavior

Hermes unexpectedly edits its own skill definitions during ordinary tasks.

The modification happens without explicit user intent to update the skill.

As a result:

• skill prompts drift over time
• previously stable agent behavior changes unexpectedly
• future generations become inconsistent

Affected Component

CLI (interactive chat)

Messaging Platform (if gateway-related)

No response

Debug Report

Debug report uploaded:
  Report     https://paste.rs/XXDOY
  agent.log  https://dpaste.com/75NRUSQW3

Operating System

Ubuntu 24.04

Python Version

No response

Hermes Version

No response

Additional Logs / Traceback (optional)

Root Cause Analysis (optional)

No response

Proposed Fix (optional)

No response

Are you willing to submit a PR for this?

• I'd like to fix this myself and submit a PR