The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.
docs: correct copilot skill count filter and document name resolution
Brian Swiger3029b44437e7
diff --git a/README.md b/README.mdindex 564199491..182b1468a 100644--- a/README.md+++ b/README.md@@ -1887,10 +1887,23 @@ To expose the full catalog, place `skills/` where Copilot CLI looks:```bashcp -r skills .github/skills # or: copilot skill add ./skills-copilot skill list --json | jq '[.[] | select(.source=="project")] | length'+copilot skill list --json | jq '[.[] | select(.path | contains(".github/skills"))] | length'# 286```+Filter on the path rather than on `source`, because `source == "project"`+returns 290 here: the 286 from `.github/skills/` plus the 3 `.claude/commands/`+entries and the one `.agents/skills/` skill whose name does not collide with a+full-catalog entry. The other 38 `.agents/skills/` entries resolve to the same+skill names as their `.github/skills/` counterparts and are de-duplicated in+favor of `.github/skills/`.++Copilot registers a skill under the `name` in its `SKILL.md` frontmatter, not+its directory name, so catalog folders whose names differ from their frontmatter+(for example `scientific-db-pubmed-database`, which declares+`name: pubmed-database`) still load — they are just listed under the frontmatter+name.+Agents work the same way:```bash