Sae Feature Annotations

What it does

When your agent uses it

• Has SAE feature indices (e.g
• The user mentions Biohub feature annotations
• The feature viewer
• The ESMC-6B layer-60 SAE codebook

Example prompts

• “s SAE model / pooledfeatures) and wants to interpret them:”
• “which features fired on my protein and what do they correspond to”
• “annotate these feature indices”
• “/sae-feature-annotations”

Requirements

• Python 3
• A credential in ESM_API_KEY
• A credential in FORGE_TOKEN

Workflow steps

2 steps, taken from the first numbered list in SKILL.md.

1. 1The whole map, in one call — GET /esm/protein/api/v1alpha1/features returns every
2. 2One feature, in full — GET /esm/protein/api/v1alpha1/features/{feature_index}

What it can do on your machine

Read from SKILL.md and the folder at commit fa8c941. It shows what the files ask for, not the result of running them.

• Tool permissions

  Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

  From allowed-tools in the SKILL.md frontmatter.

• Runs code

  Ships 1 file in scripts/ (Python), which the agent can run.

  From the folder's file list and the shell code blocks in SKILL.md.

• Network

  Hosts in commands or code, which the agent is likely to contact:

  • biohub.ai

  From URLs in SKILL.md, links to its own repository left out.

• Credentials

  Names these keys or tokens, usually read from environment variables:

  • ESM_API_KEY
  • FORGE_TOKEN

  From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Sae Feature Annotations loads about 1.5k tokens when it runs, and up to ~2.7k if it reads all its reference files. Until then it costs about 241 tokens; SKILL.md has 665 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it

~241

When it runs · the whole SKILL.md, loaded when a task matches

~1.5k

With references · SKILL.md plus every file in references/, read only if the agent opens them

~2.7k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.

SKILL.md

The full file from softnanolab/bagel at commit fa8c941, republished under its MIT licence (© softnanolab). 665 words, ~1,533 tokens.

name

sae-feature-annotations

description

Look up what a Biohub ESM-C sparse-autoencoder (SAE) feature *means* — its label, description, top-activating proteins, decoder neighbours, and activation statistics — by querying the Biohub feature-annotation API. Use this whenever the user has SAE feature indices (e.g. from `boileroom`'s `SAE` model / `pooled_features`) and wants to interpret them: "what is SAE feature 12345", "which features fired on my protein and what do they correspond to", "annotate these feature indices", "what proteins most activate this feature", "find the zinc-binding SAE feature", "decoder nearest neighbours of feature N", or any request to turn ESM-C SAE feature numbers into biology. Also trigger when the user mentions Biohub feature annotations, the feature viewer, or the ESMC-6B layer-60 SAE codebook. These annotations are only valid for the `ESMC-6B-sae-layer60-k64-codebook16384` SAE — the skill says so up front and refuses to pretend otherwise.

scripts/sae_features.py
  list   [--limit N]      # list features (map endpoint)
  search <text>           # find features whose label/description matches <text>
  get    <feature_index>  # full detail for one feature (detail endpoint)
  # shared flags: --base-url, --token, --json

© softnanolab, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

SKILL.md and 3 other files (scripts, references) in .claude/skills/sae-feature-annotations of softnanolab/bagel.

• SKILL.md
• evals/evals.json
• references/api.md
• scripts/sae_features.py

Open the folder on GitHubat commit fa8c941