hermes - 💡(How to fix) Fix CLI agent loops indefinitely: read/grep/read cycles for hours with zero progress on kanban tasks

Bug Description

When using the Hermes CLI agent to work on kanban tasks, the agent enters an infinite loop of:

• → → → → → ...

This continues for hours with zero actual progress. No files are modified. No code is written. No tests are run. The agent just reads and searches the same files repeatedly.

Steps to Reproduce

1. Load a kanban skill or receive a task with clear instructions
2. Start the agent on the task
3. Observe the agent reading files, searching, reading more files
4. After 30+ minutes, no code changes have been made
5. After 2+ hours, still looping with reads and searches

Expected Behavior

• Agent should write code, run tests, make progress within minutes
• If stuck, agent should ask for help or escalate — not loop silently
• Kanban tasks with clear instructions should result in deliverables

Actual Behavior

• Agent loops on file inspection for hours
• No , , or commands with actual build/test steps
• User has to manually intervene after hours of wasted time
• Every agent attempt results in the same read-only loop

Environment

• Platform: CLI
• Hermes version: current main (2026-05-09)
• Model: kimi-k2.6 (via kimi-for-coding), minimax-m2.7
• Task type: Kanban-driven coding tasks with clear instructions

Additional Context

This happens with every agent attempt on coding tasks. The kanban has clear instructions, but the agent never transitions from "understanding" to "doing." It seems like the agent is either:

1. Missing a "just do it" trigger in the prompt
2. Not confident enough to write code without re-reading everything
3. Lacking a timeout/max-reads guard that forces action

User expectation: "crack on" means autonomous execution — read what's needed, then write the code.