❯ i wanted to REPORT AN ISSUE I am encoutering that I am not sure others have seen.. Agent skill creation seems to lack a quality filter. It has been saving narrative summaries of one-time troubleshooting as reusable skills. For example, I was setting up the NousRouter which was a one time setup that had a small hiccup I was discussing wiith the agent. Two days later when my Danual (it's a dynamic manual I made to track changes and updates to skills and commands etc) was reviewing updates I see that it was created as a new skill in skill.md . I have many examples of this...................................Here is an example of what is occurring.
This skill called "openrouter-billing-investigate" was created after we simply looked at your logs one time to answer the question: "Why did I get a weird bill?"
**The SKILL.md file contains:**
```yaml
---
name: openrouter-billing-investigate
description: Investigate unexpected OpenRouter charges by analyzing gateway logs for fallback patterns...
---
# OpenRouter Billing Investigate
## Context
We looked at ~/Logs/gateway.log to find fallback usage on April 18.
The agent ran terminal commands to grep for "openrouter" and "billing".
## What we found
- There were 86 calls to Gemini-Flash for auxiliary tasks.
- There were 10 fallback calls to MiniMax.
- Total estimated cost was 5 cents.
```
**Why this is broken:**
1. **It's a diary entry, not a tool.** It records *what we found today*, not *how to investigate billing generally*.
2. **It will mislead the agent later.** If you ask in 6 months, "How do I check my billing?", the agent might load this file, see "86 calls to Gemini," and think that is a *hard-coded fact* about your system, rather than a one-time observation.
3. **It is static.** Real logic belongs in a script (like the `danual` scanner), not in a skill that just says "we looked at logs."
Another example is **`browser-engine-selection`**, which is just a text summary of a conversation we had about whether "cloud" or "local" browsers are better. It contains no instructions, just an opinionated summary of that specific chat.
❯ i wanted to REPORT AN ISSUE I am encoutering that I am not sure others have seen.. Agent skill creation seems to lack a quality filter. It has been saving narrative summaries of one-time troubleshooting as reusable skills. For example, I was setting up the NousRouter which was a one time setup that had a small hiccup I was discussing wiith the agent. Two days later when my Danual (it's a dynamic manual I made to track changes and updates to skills and commands etc) was reviewing updates I see that it was created as a new skill in skill.md . I have many examples of this...................................Here is an example of what is occurring.
This skill called "openrouter-billing-investigate" was created after we simply looked at your logs one time to answer the question: "Why did I get a weird bill?"