hermes - 💡(How to fix) Fix SKILLS_GUIDANCE mentions "5+ tool calls" but the main agent has no reliable way to count them — is this guidance redundant? [1 participants]

Hi, I've been reading through the skill creation mechanism and have a question about the design intent behind SKILLS_GUIDANCE.

SKILLS_GUIDANCE (in agent/prompt_builder.py) is injected into the system prompt whenever skill_manage is available:

SKILLS_GUIDANCE = (
"After completing a complex task (5+ tool calls), fixing a tricky error, "
"or discovering a non-trivial workflow, save the approach as a "
"skill with skill_manage so you can reuse it next time.\n"
...
) At the same time, there is a separate, reliable automated mechanism:

_iters_since_skill is incremented on every tool-calling iteration (run_agent.py). When it reaches _skill_nudge_interval (default 10), _spawn_background_review(review_skills=True) is triggered. A silent background agent then reviews the conversation using _SKILL_REVIEW_PROMPT and decides whether to create or update a skill. Question

The "5+ tool calls" threshold in SKILLS_GUIDANCE is presented to the main agent as a trigger condition. However, the main agent has no reliable way to count how many tool calls have been made in the current task:

There is no counter injected into the prompt telling the model "you have made N tool calls so far." The model would have to manually count tool_calls entries in the conversation history, which it does not do in practice. After context compression, the raw tool call history is replaced by a structured summary, making counting even less reliable. Given that the actual automated skill-creation path goes through _iters_since_skill → _spawn_background_review → _SKILL_REVIEW_PROMPT, is the "5+ tool calls" description in SKILLS_GUIDANCE effectively dead guidance for the main agent? Could it be simplified or removed to reduce noise in the system prompt, or is there a design intent I'm missing?

Relevant files: agent/prompt_builder.py — SKILLS_GUIDANCE run_agent.py — _iters_since_skill, _skill_nudge_interval, _spawn_background_review run_agent.py — _SKILL_REVIEW_PROMPT