pyvene-interventions

Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework. Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior.

Quality

82%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

65%

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

A richly actionable skill with executable workflows and a real, well-signaled references bundle, but it runs long and keeps reference-grade detail inline. Batch workflows also lack explicit validation checkpoints.

Suggestions

Move the inline 'Key Classes Reference', 'Component Targets', and intervention-type tables into references/api.md, leaving SKILL.md a leaner overview that links out.

Trim contextual padding (GitHub stars, full paper citation, Supported Models list, Comparison table) or relocate it to references/README.md to improve token efficiency.

Add explicit validation/verification checkpoints to the batch workflows — e.g., assert restored-probability sanity checks after the causal-tracing sweep and a held-out eval step in the IIT training loop.

Dimension	Reasoning	Score
Conciseness	The body is mostly actionable code, but includes unnecessary framing an ML engineer already knows ('pyvene is Stanford NLP's library...', paper citation, '(840+ stars)', a full comparison table, and a Supported Models list) and could be tightened.	2 / 3
Actionability	Fully executable, copy-paste-ready examples throughout — complete imports, model loading, IntervenableConfig construction, training loops, steering/generation — plus a WRONG/RIGHT 'Common Issues' section.	3 / 3
Workflow Clarity	Workflows are numbered with a checklist for causal tracing, but batch sweeps (layer×position trace, IIT training loop) lack verification/validation checkpoints, capping the score per the batch-operations guideline.	2 / 3
Progressive Disclosure	References are clearly signaled in a table pointing to real files (README.md, api.md, tutorials.md), but substantial API/class reference material remains inline in SKILL.md rather than being split out, matching the 'content that should be separate is inline' anchor.	2 / 3
	Total	9 / 12 Passed

Description

100%

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

A strong, third-person description that names the library, the domain, several concrete intervention activities, and an explicit 'Use when' trigger clause. It distinguishes itself clearly from neighboring interpretability skills.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions in the trigger clause ('causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior') alongside the specific framework, matching the 'lists multiple specific concrete actions' anchor.	3 / 3
Completeness	Explicitly answers both what ('Provides guidance for performing causal interventions on PyTorch models using pyvene's declarative intervention framework') and when ('Use when conducting causal tracing, activation patching, interchange intervention training, or testing causal hypotheses').	3 / 3
Trigger Term Quality	Natural terms an interpretability researcher would actually say ('causal tracing', 'activation patching', 'interchange intervention training', 'causal hypotheses') give good coverage rather than opaque jargon.	3 / 3
Distinctiveness Conflict Risk	The pyvene-specific declarative-intervention niche with distinct trigger terms (causal tracing, IIT, activation patching) is unlikely to fire for unrelated skills.	3 / 3
	Total	12 / 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/pyvene/SKILL.md
Commit: 52845c3

Reviewed: about 14 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.