Problem
Users who maintain skill/reference files in ~/.copilot/skills/ need them loaded at the start of every session. Today the only mechanism is a prose instruction in copilot-instructions.md saying "read these files first" — but model compliance is inconsistent. 3 out of 3 new sessions this morning required manually prompting the agent to load skills.
Proposed Solution
Add a startupFiles array in settings.json that the CLI runtime reads and injects into context automatically before the first model turn — no model compliance needed.
"startupFiles": [
"~/.copilot/skills/**/*.md",
"~/.copilot/skills/references/*.md"
]
This would guarantee domain context is always present without wasting a turn on file reads or relying on the model to follow instructions.
Impact
Any power user who builds custom skills, domain knowledge bases, or team-specific reference files hits this every single session. The current workaround is asking the agent "have you loaded skills" at the start of every session, which defeats the purpose of having persistent instructions.
Problem
Users who maintain skill/reference files in
~/.copilot/skills/need them loaded at the start of every session. Today the only mechanism is a prose instruction incopilot-instructions.mdsaying "read these files first" — but model compliance is inconsistent. 3 out of 3 new sessions this morning required manually prompting the agent to load skills.Proposed Solution
Add a
startupFilesarray insettings.jsonthat the CLI runtime reads and injects into context automatically before the first model turn — no model compliance needed.This would guarantee domain context is always present without wasting a turn on file reads or relying on the model to follow instructions.
Impact
Any power user who builds custom skills, domain knowledge bases, or team-specific reference files hits this every single session. The current workaround is asking the agent "have you loaded skills" at the start of every session, which defeats the purpose of having persistent instructions.