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 highly actionable skill with executable code across four workflows and verified, well-linked reference files. Its main weaknesses are token redundancy and a lack of explicit validation checkpoints in the batch sweep workflows.

Suggestions

Tighten the intro paragraph (it re-states the description) and trim the Supported Models list and Comparison-with-Other-Tools table to essentials to reduce token load.

Add explicit validation feedback loops to the batch sweep workflows — e.g., sanity-check that sweep results have the expected (n_layers, seq_len) shape and reasonable value ranges before interpreting hotspots, or verify logit diff against a clean-vs-corrupted baseline.

Move the inline API reference tables (Intervention Types, Component Targets, Key Classes Reference) into references/api.md to eliminate duplication and keep the body a lean overview that points outward.

Dimension	Reasoning	Score
Conciseness	The body is mostly task-oriented executable code, but the opening paragraph restates the description and the Supported Models list and Comparison-with-Other-Tools table add tokens that could be trimmed.	2 / 3
Actionability	Each workflow ships copy-paste-ready, executable code (IntervenableModel setup, causal-trace sweep, patching loop, IIT training, steering, save/load) plus concrete tables of intervention types and components.	3 / 3
Workflow Clarity	Workflows have clear numbered steps and Workflow 1 includes a checklist, but the batch sweeps (causal tracing, patching) lack explicit validation/checkpoint feedback loops, capping clarity at 2 per the batch-operations guideline.	2 / 3
Progressive Disclosure	References are well-signaled one level deep and the three bundle files exist and are linked, but inline API tables (Intervention Types, Component Targets, Key Classes Reference) duplicate material in references/api.md — content that should be separate is inline.	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 clearly states the capability and gives explicit 'Use when' triggers covering the library's main intervention tasks. It is specific, well-triggered, and unlikely to conflict with other skills.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete actions — 'causal tracing, activation patching, interchange intervention training, or testing causal hypotheses about model behavior' — matching the multi-action 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...').	3 / 3
Trigger Term Quality	'causal tracing', 'activation patching', and 'interchange intervention training' are the natural terms a user would say when they need this skill, with good variation coverage.	3 / 3
Distinctiveness Conflict Risk	The pyvene causal-intervention niche with these specific triggers is clearly distinguishable and 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: OpenLAIR/dr-claw
Path: skills/mechanistic-interpretability/pyvene/SKILL.md
Commit: 6da7f7c

Reviewed: about 7 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.