Borrowing it
Nothing to install: this file belongs to softnanolab/bagel. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/softnanolab/bagel/main/.claude/skills/sae-feature-annotations/SKILL.mdgit clone --depth 1 https://github.com/softnanolab/bagelWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/softnanolab/bagel/sae-feature-annotations)<a href="https://agentmods.dev/skills/softnanolab/bagel/sae-feature-annotations"><img src="https://agentmods.dev/badge/skills/softnanolab/bagel/sae-feature-annotations/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/softnanolab/bagel/sae-feature-annotations"><img src="https://agentmods.dev/badge/skills/softnanolab/bagel/sae-feature-annotations.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00230 | $0.01550 |
| Opus 5.5 | $0.00092 | $0.00620 |
| Sonnet 5.5 | $0.00046 | $0.00310 |
| Haiku 4.5 | $0.00023 | $0.00155 |
Grade A, and why
sae-feature-annotations scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 116 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Annotating ESM-C SAE features from Biohub
This skill turns SAE feature indices into biology. A sparse autoencoder trained on ESM-C representations decomposes each residue into a sparse set of interpretable features (directions in a codebook). Biohub publishes an annotation for each feature — a human label, top-activating proteins, decoder neighbours, activation stats — and this skill fetches and presents them by querying the Biohub API live.
The one thing to say first, every time
These annotations are only valid for the SAE ESMC-6B-sae-layer60-k64-codebook16384
— ESM-C 6B, transformer layer 60, TopK k=64, codebook of 2**14 = 16384 features. This is
the SAE from Language Modeling Materializes a World Model of Protein Biology (Biohub,
2026), and the default Forge SAE that boileroom's SAE model uses.
A feature_index (0…16383) means something completely different under any other layer, k,
or codebook. So before interpreting anything, confirm the user produced their features
with this exact SAE (in boileroom that is the default feature_source="forge" path,
i.e. forge_sae_model = "esmc-6b-2024-12-sae-layer60-k64-codebook16384"). If they used a
local 300M/600M SAE or a different layer, tell them these labels do not apply and stop — do
not hand them annotations that describe a different feature basis.
Lead with this. Do not bury it under the results.
One source: the live Biohub API
There is no bundled offline table. The two Biohub annotation endpoints are public
reads, so the skill always queries them live. An API key is optional: if the user has
one (ESM_API_KEY / FORGE_TOKEN, the same credential boileroom's Forge backend uses)
it is sent; if not, the request is made with an anonymous placeholder token, which is
enough for the annotation endpoints. Base URL defaults to https://biohub.ai.
If a deployment ever enforces auth and the anonymous call is rejected, the fix is to set
ESM_API_KEY. That is the only case where a key matters for reading annotations —
producing the features themselves is a different tool (see the last note).
What ships with it
3 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2mo ago First seen · 116 lines · 230 tokens per session scan A 91174e0fb024
sae-feature-annotations is a skill published in the GitHub repository softnanolab/bagel (147 stars, last pushed 10d ago), licensed MIT. It adds 230 tokens to every session and 1,550 once invoked, about $0.0009 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
sae-feature-annotations compared with skills from other repositories
| Mod | Type | Stars | Last push | Licence | Scan | Tokens | Works with |
|---|---|---|---|---|---|---|---|
| sae-feature-annotationssoftnanolab/bagel | Skill | 147 | 10d ago | MIT | A | 230 | Claude Code |
| instrument-data-to-allotropeanthropics/knowledge-work-plugins | Skill | 35k | yesterday | Apache-2.0 | A | 123 | Claude Code |
| matlabK-Dense-AI/scientific-agent-skills | Skill | 56k | 6d ago | MIT | A | 44 | Claude Code |
| exploratory-data-analysisK-Dense-AI/scientific-agent-skills | Skill | 56k | 6d ago | MIT | A | 83 | Claude Code |
| phylogeneticsK-Dense-AI/scientific-agent-skills | Skill | 56k | 6d ago | MIT | A | 71 | Claude Code, Codex |
| research-engineerdavila7/claude-code-templates | Skill | 33k | today | MIT | A | 43 | Claude Code, Codex |
| mapping-to-snomedmaziyarpanahi/openmed | Skill | 5.8k | today | Apache-2.0 | A | 205 | Claude Code |