CtrlK
BlogDocsLog inGet started
Tessl Logo

rwkv-architecture

RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.

52

Quality

57%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

SecuritybySnyk

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./01-model-architecture/rwkv/SKILL.md
SKILL.md
Quality
Evals
Security

Quality

Content

57%

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is well-structured with strong executable inference examples and excellent progressive disclosure via real, well-signaled reference files. Its weaknesses are verbosity from redundant advantage lists and conceptual commentary, an incomplete fine-tuning example, and missing validation/feedback loops in the workflows.

Suggestions

Tighten the body by removing the redundant "Key advantages" list and the conceptual attention/KV-cache commentary in Workflow 4, which Claude already knows.

Make the fine-tuning workflow fully executable: define the dataloader and use the correct RWKV loading API (or mark it explicitly as illustrative pseudocode).

Add explicit validation/verification checkpoints to the batch workflows (e.g., check loss/grad-norm during training, verify state continuity after chunked long-context processing).

DimensionReasoningScore

Conciseness

Mostly efficient with substantial executable code, but it includes unnecessary explanation Claude already knows (the O(n^2) attention / KV-cache commentary in Workflow 4, the "1,000,000x more efficient" marketing claims) and a redundant "Key advantages" list that repeats facts already in the description and Quick start.

2 / 3

Actionability

The inference workflows (Quick start, streaming, long context) are concrete and copy-paste ready, but the fine-tuning example is incomplete (undefined `train_dataloader`, `RWKV(config)` constructor that does not match the library's `RWKV(model=path)` API shown elsewhere), so not all guidance is fully executable.

2 / 3

Workflow Clarity

Workflows 1-4 are clearly labeled with sequenced, commented steps, but none include explicit validation checkpoints or validate->fix->retry feedback loops; given batch operations like fine-tuning are present, the rubric caps workflow clarity at 2.

2 / 3

Progressive Disclosure

The Advanced topics section clearly signals one-level-deep references (architecture-details.md, state-management.md, rwkv7.md), each annotated with what it contains, and all three referenced files exist, keeping the overview in SKILL.md while details live in references.

3 / 3

Total

9

/

12

Passed

Description

57%

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific and occupies a distinct niche, but it reads as a spec sheet of RWKV's properties rather than a statement of what the skill does, and it entirely lacks a "Use when..." trigger clause. Adding explicit use-when guidance and action verbs would raise both completeness and specificity.

Suggestions

Add an explicit "Use when..." clause naming the situations that should trigger this skill (e.g., working with RWKV models, needing linear-time/infinite-context inference).

Replace property statements with action verbs describing what the skill does (e.g., "Load, run, and fine-tune RWKV models...") to improve specificity.

Add common user-facing trigger variations such as "RWKV model", "linear attention/RNN", and "train RWKV" alongside the existing jargon.

DimensionReasoningScore

Specificity

The description lists concrete technical properties ("O(n) inference", "infinite context", "no KV cache", "Train like GPT (parallel), infer like RNN (sequential)") but these are characteristics of the architecture rather than concrete actions the skill performs, so it names the domain and some capabilities without comprehensive actions.

2 / 3

Completeness

It states what RWKV is (its properties and provenance) but has no "Use when..." clause or equivalent trigger guidance, so per the rubric cap the "when" is missing and completeness is capped at 2.

2 / 3

Trigger Term Quality

It includes relevant natural terms a user might say ("RNN", "Transformer", "RWKV-7", "inference", "GPT", "RNN") but is heavy on jargon ("O(n) inference", "no KV cache", "Receptance Weighted Key Value") and misses common variations like "RWKV model", "linear attention", or "train RWKV".

2 / 3

Distinctiveness Conflict Risk

"RNN+Transformer hybrid", "RWKV-7", and "no KV cache" describe a clear, narrow niche (the RWKV architecture) that is unlikely to conflict with or trigger for unrelated skills.

3 / 3

Total

9

/

12

Passed

Validation

93%

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation15 / 16 Passed

Validation for skill structure

CriteriaDescriptionResult

frontmatter_unknown_keys

Unknown frontmatter key(s) found; consider removing or moving to metadata

Warning

Total

15

/

16

Passed

Repository
Orchestra-Research/AI-Research-SKILLs
Reviewed

Table of Contents

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.