@@ -4,7 +4,7 @@ Generated from the registry, the skill sidecars and the two maintained
44matrices. ** Do not edit by hand** ; regenerate after any registry, skill,
55matrix or frozen-plan change.
66
7- - Input digest: ` sha256:d162aef1743d44598b08f43e1b458c8aefcb523912b1622c78ea4f1ca714d20f `
7+ - Input digest: ` sha256:0a0e702358904ca514516773126c8965b7741ef4943c7ec8c7c3a2600bcaa1db `
88- Regenerate: ` python tools/build_skill_coverage.py `
99- Verify without writing: ` python tools/build_skill_coverage.py --check `
1010
@@ -31,17 +31,17 @@ matrix: every field is copied from an owner that already maintains it.
3131
3232| Measure | Value |
3333| ---| ---|
34- | Registered skill ids | 68 |
35- | By category | frameworks 3, hardware 22, methods 18, repositories 19 , workflows 6 |
36- | By backend | ascend 14, common 50 , cuda 4 |
34+ | Registered skill ids | 69 |
35+ | By category | frameworks 3, hardware 22, methods 18, repositories 20 , workflows 6 |
36+ | By backend | ascend 14, common 51 , cuda 4 |
3737| Recorded on a numbered SearchSpace resource layer | 47 |
38- | Recorded only in a supplementary guidance table (†) | 8 |
38+ | Recorded only in a supplementary guidance table (†) | 9 |
3939| Recorded on two or more resource layers | 24 |
40- | Recorded in the Reference Matrix | 62 |
41- | With a ` metadata.json ` sidecar | 63 |
40+ | Recorded in the Reference Matrix | 63 |
41+ | With a ` metadata.json ` sidecar | 64 |
4242| Selected by a frozen evaluation plan | 2 |
43- | ` SKILL.md ` documents in the library | 104 |
44- | Unregistered documents (excluded from every count above) | 36 |
43+ | ` SKILL.md ` documents in the library | 108 |
44+ | Unregistered documents (excluded from every count above) | 39 |
4545
4646A complete inventory is not a coverage claim. Every registered row below
4747has an entry; that says the listing is exhaustive, not that inference
@@ -82,7 +82,7 @@ optimization is fully covered or that any procedure has measured benefit.
8282| [ diffusers-invocation-state] ( ../src/inferrsi/data/skill_library/frameworks/diffusers/diffusers-invocation-state/SKILL.md ) | common | Trace Diffusers pipeline-wrapper and scheduler invocation state when repeated calls, from_pipe reuse or interruption ... | 2 RAM/Host † guidance: Diffusers invocation guidance (source-derived) | Diffusers invocation source extraction (huggingface/diffusers @ ` aea7b70be0f594bcdc15a6f2677ee2ca85e49e8e ` ...) | candidate · not_run · 1 readset(s) | no-eval |
8383| [ lazy-evaluation-stream-lifetimes] ( ../src/inferrsi/data/skill_library/frameworks/mlx/lazy-evaluation-stream-lifetimes/SKILL.md ) | common | Diagnose MLX lazy evaluation, cross-stream ordering and array-buffer lifetimes when timing, memory use or asynchronous ... | 1 Accelerator † guidance: MLX framework guidance (source-derived) | MLX framework source extraction (ml-explore/mlx @ ` bff10220c742907f0c68e55ff174aaea56f74e3d ` (MIT)) | candidate · not_run · 1 readset(s) | no-eval |
8484
85- ## Registered: Repositories (19 )
85+ ## Registered: Repositories (20 )
8686
8787| Skill | Backend | Declared task | SearchSpace layer / stage | Reference Matrix | State | Gaps |
8888| ---| ---| ---| ---| ---| ---| ---|
@@ -102,6 +102,7 @@ optimization is fully covered or that any procedure has measured benefit.
102102| [ husky] ( ../src/inferrsi/data/skill_library/repositories/husky/SKILL.md ) | common | Review Husky's official-document claims as research hypotheses, separating author reports, implementation evidence ... | — | Husky document-only research adaptation (Official https://husky.underdog.ai/ ; displayed2026-09-20 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
103103| [ kv-tier-management] ( ../src/inferrsi/data/skill_library/repositories/lmcache/kv-tier-management/SKILL.md ) | common | Manage runtime KV cache tiers (GPU → CPU RAM → local disk) for LLM serving: decide when KV is stored, retained ... | 2 RAM/Host · Configure / Optimize<br >3 SSD · Optimize | 1 · Systems sources (LMCache/LMCache @ c3b875a) | candidate · not_run · 0 readset(s) | no-eval |
104104| [ lily] ( ../src/inferrsi/data/skill_library/repositories/lily/SKILL.md ) | common | Investigate Lily's exact Qwen3.6-35B-A3B MLX-affine-Q4 contract, Metal state and kernels, strict greedy text API ... | — | Lily portable AREX extraction (perplexityai/pplx-garden ` lily/ ` @ `1ed972ed3f0bd5616c997c9507c25616c6 ...) | candidate · not_run · 2 readset(s) | no-SearchSpace-layer, no-eval |
105+ | [ mistral-rs] ( ../src/inferrsi/data/skill_library/repositories/mistral-rs/SKILL.md ) | common | Investigate mistral.rs Metal paged-attention decode layouts, prefill coordinates and workspace, actual-model prewarm ... | 1 Accelerator † guidance: mistral.rs Metal paged-attention source contracts<br >2 RAM/Host † guidance: mistral.rs Metal paged-attention source contracts | mistral.rs Metal paged-attention extraction (EricLBuehler/mistral.rs @ ` 9459891e7697238225885518119682ac8209a612 ` ...) | candidate · not_run · 2 readset(s) | no-eval |
105106| [ project-maya] ( ../src/inferrsi/data/skill_library/repositories/project-maya/SKILL.md ) | common | Investigate Project Maya model-family dispatch, representation and memory tiers, API versus native state, speculative ... | 1 Accelerator † guidance: Project Maya model-specific source contracts<br >2 RAM/Host † guidance: Project Maya model-specific source contracts<br >3 SSD † guidance: Project Maya model-specific source contracts | Project Maya source and CPU-slice extraction (mw00/project-maya @ ` 249f9a87d1a9d72bec2b684d704b0aa9296b1c9c ` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
106107| [ splash] ( ../src/inferrsi/data/skill_library/repositories/splash/SKILL.md ) | common | Investigate Splash model-format compatibility, automatic memory planning, Metal and ANE execution, DFlash2 ... | — | Splash portable AREX extraction (incoai/splash @ ` 39a97385f8560cf8a93972cec9b0a963c9195250 ` ; AREX ...) | candidate · partial_cpu_only · 2 readset(s) | no-SearchSpace-layer, no-eval |
107108| [ vllm-metal] ( ../src/inferrsi/data/skill_library/repositories/vllm-metal/SKILL.md ) | common | Investigate vllm-metal Apple inference through platform and native-ABI selection, paged KV/offload ownership ... | 1 Accelerator † guidance: vLLM Metal selected source contracts<br >2 RAM/Host † guidance: vLLM Metal selected source contracts<br >3 SSD † guidance: vLLM Metal selected source contracts | vLLM Metal portable AREX source and CPU extraction (vllm-project/vllm-metal @ ` e7df7a6ff215f99c7842552d230fdbc709835934 ` ...) | candidate · partial_cpu_only · 2 readset(s) | no-eval |
@@ -142,7 +143,7 @@ optimization is fully covered or that any procedure has measured benefit.
142143
143144## Documents outside the registry
144145
145- 36 ` SKILL.md ` documents are not registered, so they have
146+ 39 ` SKILL.md ` documents are not registered, so they have
146147no stable id and no row above. They are engine entrypoints, their
147148sub-skills, and the research router. Counting them would inflate coverage:
148149an entrypoint routes to procedures below it rather than adding one.
@@ -155,6 +156,9 @@ an entrypoint routes to procedures below it rather than adding one.
155156| ` repositories/lily/sub-skills/model-runtime ` | sub-skill |
156157| ` repositories/lily/sub-skills/qualification ` | sub-skill |
157158| ` repositories/lily/sub-skills/serving-cache ` | sub-skill |
159+ | ` repositories/mistral-rs/sub-skills/decode-contract ` | sub-skill |
160+ | ` repositories/mistral-rs/sub-skills/prefill-contract ` | sub-skill |
161+ | ` repositories/mistral-rs/sub-skills/qualification ` | sub-skill |
158162| ` repositories/project-maya/sub-skills/model-runtime ` | sub-skill |
159163| ` repositories/project-maya/sub-skills/qualification ` | sub-skill |
160164| ` repositories/project-maya/sub-skills/serving-state ` | sub-skill |
@@ -188,7 +192,7 @@ an entrypoint routes to procedures below it rather than adding one.
188192
189193## Integrity checks
190194
191- - Registered ids resolve to an existing ` SKILL.md ` : 68 of 68 .
195+ - Registered ids resolve to an existing ` SKILL.md ` : 69 of 69 .
192196- Duplicate ids or paths: none; ` skill_catalog ` rejects them before this view is built.
193197- Skill links in the two matrices that do not resolve: 0.
194198- Registered skills without a ` metadata.json ` sidecar: 5.
@@ -200,7 +204,7 @@ an entrypoint routes to procedures below it rather than adding one.
200204| ` no-SearchSpace-layer ` | 13: ascend-video-diffusion-optimization, audit-recurrent-kv-finalization, cite-check-downstream-pr, competitor-scan, draft-acceptance-issue, dwarfstar, husky, inference-benchmark-design, lily, npu-skill-factory, repository-evolution-analysis, skill-evolution-playbook, splash |
201205| ` no-Reference-Matrix-row ` | 6: ascend-video-diffusion-optimization, cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence |
202206| ` no-sidecar ` | 5: cite-check-downstream-pr, claim-contract, competitor-scan, draft-acceptance-issue, rsi-evidence |
203- | ` no-eval ` | 66 : as listed in the Gaps column above |
207+ | ` no-eval ` | 67 : as listed in the Gaps column above |
204208
205209These are inventory gaps, not defects. A workflow skill may legitimately
206210have no hardware resource layer; a candidate may legitimately be untested.
0 commit comments