Agents almost always fail to use explicit skills specified in prompt for skil...

Agents almost always fail to use explicit skills specified in prompt for skills that are scoped to repository.

They often find the skill and use it if i make them look for it first in the repo as a regular file. Otherwise they always say that the runtime’s loaded skill registry did not include this repo skill.

---
| Field | Value |
| --- | --- |
| App version | 1.1.9 |
| OS | Windows 10.0.26100 |
| Theme | GitHub |
| Path | /chat |
| Tenure | Week 13 |

Reproduce the behavior described for repository-scoped skills in the /chat experience on Windows 10. Start by tracing how explicit skills are discovered and how the runtime’s loaded skill registry is populated; no source files or tests are named in the report. Done means an explicitly requested repository skill is found and used without first asking the agent to locate it as a regular file.

Written by the indexing model from the issue text.