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.

Quality

70%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Passed

No findings from the security scan

Fix and improve this skill with Tessl

tessl review fix ./backend/cli/skills/ml-training/rwkv/SKILL.md

Quality

Content

72%

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

The content is highly actionable with executable examples and good progressive disclosure into real reference files, but it carries some pedagogical padding and lacks validation feedback loops for its destructive/batch training workflows.

Suggestions

Trim the explanatory comparison comments in Workflow 4 and the 'When to use vs alternatives' prose that re-explains concepts Claude already knows.

Add explicit validation checkpoints to the fine-tuning/training workflows (e.g., verify checkpoint loads, validate loss/grad-norm before scaling to 8 GPUs, a validate→fix→retry loop).

Dimension	Reasoning	Score
Conciseness	The body is mostly lean executable code with brief framing, but the 'When to use vs alternatives' section and the pedagogical comparison comments in Workflow 4 (e.g., '# 1,000,000× more efficient!') re-explain Transformer-vs-RNN concepts Claude already knows and could be trimmed.	2 / 3
Actionability	Throughout, it provides fully executable, copy-paste-ready code and commands: pip install lines, forward() calls, a training config, and concrete OOM/slow-inference fixes — matching the fully-executable anchor.	3 / 3
Workflow Clarity	Several workflows are sequenced, but the risky/batch operations (fine-tuning, multi-GPU training) lack validation checkpoints or feedback loops, and Workflow 4 is conceptual rather than a real sequence — per the notes this caps workflow clarity at 2.	2 / 3
Progressive Disclosure	The SKILL.md body is a clear overview with three well-signaled, one-level-deep references (architecture-details.md, state-management.md, rwkv7.md), all of which exist as real bundle files — matching the clear-overview anchor.	3 / 3
	Total	10 / 12 Passed

Description

67%

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 distinctive about RWKV's capabilities but is written as a feature list rather than as invocation guidance, omitting any 'when to use' trigger clause. Voice is correctly third person throughout.

Suggestions

Add an explicit 'Use when...' clause naming natural user phrases (e.g., 'Use when working with RWKV models, linear-attention/RNN-Transformer hybrids, or infinite-context inference').

Soften technical jargon with a few everyday trigger terms users would actually say, such as 'RWKV models', 'linear-complexity attention', or 'long-context generation'.

Dimension	Reasoning	Score
Specificity	Names multiple concrete capabilities and properties such as 'O(n) inference', 'Train like GPT (parallel), infer like RNN (sequential)', 'no KV cache', and 'Models up to 14B parameters', matching the multiple-specific-actions anchor.	3 / 3
Completeness	It clearly answers 'what does this do' but contains no 'Use when...' or equivalent explicit 'when' guidance, which per the guidelines caps completeness at 2.	2 / 3
Trigger Term Quality	It includes relevant domain terms (RNN, Transformer, RWKV, KV cache, GPT) but they are mostly proper-noun jargon; it lacks the common natural-language variations a user would actually say when requesting this skill.	2 / 3
Distinctiveness Conflict Risk	The RWKV-specific niche ('Linux Foundation AI project', 'RWKV-7', 'no KV cache') is a clear, distinct trigger unlikely to fire for any other skill.	3 / 3
	Total	10 / 12 Passed

Validation

93%

Warnings & errors only

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

Validation — 15 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning

	Total	15 / 16 Passed

Repository: synthetic-sciences/openscience
Path: backend/cli/skills/ml-training/rwkv/SKILL.md
Commit: 52845c3

Reviewed: about 13 hours ago

Table of Contents

Discovery Implementation Validation

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.