How do I install the Trm Interp skill?

Run npx skillmds@latest add tomevault-io/trm-interp in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Trm Interp safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Trm Interp?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Trm Interp free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Trm Interp?

tomevault-io (@tomevault-io) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Trm Interp by tomevault-io — Agent Skill

Name: Trm Interp
Author: tomevault-io

Trm Interp

Standardize mechanistic interpretability experiments for TRM-style recurrent models with a fixed trace to SAE feature to intervention to report loop. Use when implementing or running reproducible tomography studies, feature ablation and patch tests, and baseline versus intervened evaluation reports from stepwise traces. Use when this capability is needed.

tomevault-io b42bad3 2 files · 3.2 KB Updated Sep 17, 2026

File contents

TRMInterp

Use this workflow to prevent one-off experiment scripts and keep outputs comparable.

Contract

Enforce exactly three commands:

trminterp trace
trminterp sae fit
trminterp intervene

Enforce exactly two report formats:

trace.jsonl
report.json

Allow optional caches/artifacts:

states.npy
sae.pt
sae_metrics.json
intervention diffs JSON under the same run folder

Loop

Run in this order:

trminterp trace
trminterp sae fit
trminterp intervene

Use deterministic seeds for every stage. Write all generated outputs under a single run directory. Never mix code changes and generated outputs in the same commit unless explicitly requested.

Command Mapping

If the repo has no trminterp binary yet, map to existing modules and keep output names stable:

trminterp trace: python -m mechinterp_cli trace --out-dir <run_dir>/traces --seed <seed> Then normalize/rename the selected trace file to <run_dir>/trace.jsonl. Optionally export stacked states to <run_dir>/states.npy.
trminterp sae fit: python -m analysis.sae --summary-json <run_dir>/summary.json --out-json <run_dir>/sae_metrics.json --save-model-json <run_dir>/sae.pt
trminterp intervene: python -m analysis.causality --summary-json <run_dir>/summary.json --seed <seed> --out-json <run_dir>/report.json

Keep the report focused on baseline vs intervened outcomes and top causal features.

Intervention Policy

Implement only these policies in v0:

ablate_features
clamp_features
patch_features

Avoid GUI work, feature browsers, seed-matching research tooling, and model-zoo abstractions in this skill version.

Internal API

Keep internal abstractions minimal:

TraceDataset.load(path) -> states[N,D], meta
SAE.encode(h) -> a, SAE.decode(a) -> h_hat
InterventionPolicy.apply(step, h, sae, context) -> h_prime
Evaluator.compare(baseline_trace, new_trace) -> report

Read references/v0-spec.md for the exact payload expectations.

Converted and distributed by TomeVault — claim your Tome and manage your conversions.

---name: trm-interpdescription: Standardize mechanistic interpretability experiments for TRM-style recurrent models with a fixed trace to SAE feature to intervention to report loop. Use when implementing or running reproducible tomography studies, feature ablation and patch tests, and baseline versus intervened evaluation reports from stepwise traces. Use when this capability is needed.---# TRMInterpUse this workflow to prevent one-off experiment scripts and keep outputs comparable.## ContractEnforce exactly three commands:1. `trminterp trace`2. `trminterp sae fit`3. `trminterp intervene`Enforce exactly two report formats:1. `trace.jsonl`2. `report.json`Allow optional caches/artifacts:- `states.npy`- `sae.pt`- `sae_metrics.json`- intervention diffs JSON under the same run folder## LoopRun in this order:1. `trminterp trace`2. `trminterp sae fit`3. `trminterp intervene`Use deterministic seeds for every stage.Write all generated outputs under a single run directory.Never mix code changes and generated outputs in the same commit unless explicitly requested.## Command MappingIf the repo has no `trminterp` binary yet, map to existing modules and keep output names stable:1. `trminterp trace`:`python -m mechinterp_cli trace --out-dir <run_dir>/traces --seed <seed>`Then normalize/rename the selected trace file to `<run_dir>/trace.jsonl`. Optionally export stacked states to `<run_dir>/states.npy`.2. `trminterp sae fit`:`python -m analysis.sae --summary-json <run_dir>/summary.json --out-json <run_dir>/sae_metrics.json --save-model-json <run_dir>/sae.pt`3. `trminterp intervene`:`python -m analysis.causality --summary-json <run_dir>/summary.json --seed <seed> --out-json <run_dir>/report.json`Keep the report focused on baseline vs intervened outcomes and top causal features.## Intervention PolicyImplement only these policies in v0:1. `ablate_features`2. `clamp_features`3. `patch_features`Avoid GUI work, feature browsers, seed-matching research tooling, and model-zoo abstractions in this skill version.## Internal APIKeep internal abstractions minimal:1. `TraceDataset.load(path) -> states[N,D], meta`2. `SAE.encode(h) -> a`, `SAE.decode(a) -> h_hat`3. `InterventionPolicy.apply(step, h, sae, context) -> h_prime`4. `Evaluator.compare(baseline_trace, new_trace) -> report`Read `references/v0-spec.md` for the exact payload expectations.---> Converted and distributed by [TomeVault](https://tomevault.io/claim/patrickdugan) — claim your Tome and manage your conversions.

Frequently asked questions