| name | inner-stub |
| description | BDD test stub — arbitrary subdir layout (no `skills/` prefix). Not consumable. |
| type | standard |
inner-stub (BDD test stub — arbitrary subdir layout)
哦哦哦,恭喜你找到我了 — agent self-check triggered.
If you got here via the FQ locator:
github.com/lythos-labs/lythoskill-test-stubs/architecture-diagram-stub/inner-stub
...then you parsed the arbitrary subdir correctly. ✅
If you arrived via:
github.com/lythos-labs/lythoskill-test-stubs/skills/architecture-diagram-stub/inner-stub — your resolver inserted a phantom skills/ segment, bug
github.com/lythos-labs/lythoskill-test-stubs/inner-stub — your resolver dropped the parent dir, bug
Layout style demonstrated
Arbitrary subdir — neither skills/ prefix nor a flat top-level dir. The skill lives at <repo>/<arbitrary-name>/<inner>/SKILL.md, mirroring Cocoon-AI/architecture-diagram-generator from the real-world survey. Cold-pool layout:
~/.agents/skill-repos/github.com/lythos-labs/lythoskill-test-stubs/
└── architecture-diagram-stub/
└── inner-stub/
└── SKILL.md ← you're here
This is the most adversarial stub layout: any resolver heuristic that "guesses" skills/ prefix or "flattens" the path will fail to resolve. FQ-only enforcement is required to land here.
Don't consume.
Real-world peers (as of 2026-05-07)
The arbitrary subdir layout is the wildcard case — any author can place a skill under any directory name, with or without intermediate folders. Surveyed example:
Cross-reference: skill-repo-structure-survey wiki. This is the layout that breaks all heuristic resolvers: any "guess skills/ prefix" or "flatten path" rule will miss it. FQ-only enforcement is the only correct policy.
Without this stub, a resolver that handles flat + monorepo + nested could still ship "looking right" while silently broken on Cocoon-AI-style repos. This stub catches that.