Skills token limit impact on context window

[saw235]

Issue Description

When attempting to use a skill, it may not be visible in the system prompt due to token limits, even though the skill loads successfully.

Steps to Reproduce

1. Have multiple skills installed (e.g., 49 skills)
2. Invoke a skill using the skill tool
3. Skill loads successfully with message: Skill "skill-name" loaded successfully. Follow the instructions in the skill context.
4. However, the skill is not visible in the <available_skills> section of the system prompt

Current Behavior

• Only a subset of skills are shown in the system prompt (31 of 49 in this case)
• Skills that are hidden due to token limits can still be invoked
• This creates confusion as the AI cannot see the skill instructions until after invocation

Expected Behavior

All available skills should be accessible to the AI, or there should be a better mechanism for handling skill discovery and loading when many skills are installed.

Not sure why there are such limitations to the number of skills. But I would expect that since the context for them is significantly less than MCP, they should be able to be properly visible to the cli tool.

[rosspeili]

We hit the same wall with stacked skills eating context, Skillware’s monitoring/token_limiter is a model-agnostic middleware skill that returns CONTINUE / WARN / FORCE_TERMINATE from cumulative token counts so your host loop can bail before the window’s gone (pip install skillware to test), the next version would match limits against given roi, eg. you have Xk tokens to deliver Y, if one of two is not true, abort, say create a script to remove the background of an image under 20k tokens. If it tries to spend more than 20k tokens, terminate; if it delivers something, but the deliverable is not in parse with the initial intent, eg. bg was not removed correctly, also fail.

[copilot-cli-bot]

This appears to be resolved as of GitHub Copilot CLI v1.0.86. Please update to that release or a newer stable version. We're closing this as completed. If it still happens, please comment with your version and reproduction steps.