Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add jmschrei/ledidi --skill datagit clone --depth 1 https://github.com/jmschrei/ledidiWrote 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/jmschrei/ledidi/data)<a href="https://agentmods.dev/skills/jmschrei/ledidi/data"><img src="https://agentmods.dev/badge/skills/jmschrei/ledidi/data/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/jmschrei/ledidi/data"><img src="https://agentmods.dev/badge/skills/jmschrei/ledidi/data.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 | $0.00126 | $0.01771 |
| Opus 5.5 | $0.00050 | $0.00708 |
| Sonnet 5.5 | $0.00025 | $0.00354 |
| Haiku 4.5 | $0.00013 | $0.00177 |
Grade A, and why
ledidi 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 1mo 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 — 117 lines — stays where its author put it; the contents beside it link to each section on GitHub.
ledidi
ledidi inverts the usual training loop: the oracle model is frozen and the
data is optimized. It learns a continuous weight matrix, samples one-hot edits
to a template sequence from a Gumbel-softmax, and pushes those edits until the
oracle predicts what you asked for — while an input loss keeps the number of
edits small. Any differentiable PyTorch model that maps a one-hot sequence to a
prediction becomes a sequence editor.
This skill is a router. Each topic below has a reference file with exact signatures and footguns. Read the relevant reference file before writing code — do not rely on memory of the API. Several of ledidi's failure modes are silent: a plausible-looking argument can broadcast against the wrong shape, protect the region you meant to edit, or prune against the wrong objective, with no exception raised.
tangermeme is a hard dependency of ledidi (it supplies the input validation),
and it owns everything that happens around a design — one-hot encoding, FASTA
and loci I/O, attributions, motif scanning, logo plotting. Install its skill
too (tangermeme-install-skills) and consult it for those steps rather than
reinventing them here.
Is ledidi even the right tool? ledidi is gradient-based and finds small,
targeted edits to an existing template. For discrete design — implanting motifs
from a library, screening random candidates, greedy or beam substitution — use
tangermeme.design instead (screen, greedy_substitution, beam_substitution,
greedy_marginalize); its skill's references/design.md covers them. Note that
tangermeme.design requires a per-candidate loss (reduction='none') to rank
edits, which is the opposite of ledidi's requirement that the output loss
return a scalar — do not carry that habit across.
Read these first
- The objective,
references/objective.md— every design isoutput_loss(y_hat, y_bar) + l * input_loss, where the input loss is the mean number of edits per sequence.lis the exchange rate between "hit the target" and "make few edits", and it is the knob you will actually tune. Also covers what theverboselog lines mean and why the returned design is the best-scoring iterate rather than the last one.
What ships with it
19 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.
- references/catalogs-and-repeats.md 3.7 KB
- references/custom-losses.md 9.3 KB
- references/designer-object.md 5.6 KB
- references/first-design.md 3.5 KB
- references/initial-weights.md 5.7 KB
- references/inpainting.md 4.2 KB
- references/io-and-validation.md 7.2 KB
- references/masks.md 4.3 KB
- references/memory-and-oom.md 9.8 KB
- references/multi-task-models.md 5.2 KB
- references/multiple-models.md 3.8 KB
- references/objective.md 7.4 KB
- references/oracle-contract.md 4.5 KB
- references/pipeline.md 5.9 KB
- references/plotting.md 4.4 KB
- references/pruning.md 4.1 KB
- references/receptive-field.md 5.4 KB
- references/reproducibility.md 4.5 KB
- references/validating-designs.md 6.5 KB
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.
- 1mo ago Changed 1013ab068cc8
- 1mo ago Changed 7a27eb52d2fa
- 1mo ago First seen · 117 lines · 126 tokens per session scan A c16420107429
ledidi is a skill published in the GitHub repository jmschrei/ledidi (117 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 126 tokens to every session and 1,771 once invoked, about $0.0005 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.
Other skills, from other repositories
arboreto
Infers candidate gene regulatory networks from bulk or single-cell expression data using AertsLab Arboreto GRNBoost2 and GENIE3. Use for transcription factor-target association ranking, compatible Dask execution, sparse expression inputs, and network stability checks.
deepspot-m
Generates transcriptome-wide virtual spatial transcriptomics from H&E histology with DeepSpot-M. Used for predicted log1p-CPM expression from 224x224 tiles at about 20x, querying the released protein-coding gene panel by symbol, and whole-slide prediction after resolution-aware tiling with histolab.
torchdrug
Builds and troubleshoots TorchDrug 0.2.1 workflows for molecular graphs, property prediction, self-supervised pretraining, molecule generation, retrosynthesis, protein representation learning, and knowledge graph reasoning. Use when code imports torchdrug or needs its datasets, models, tasks, or Engine.
pyhealth
Builds and validates PyHealth clinical machine-learning pipelines for EHR, signals, imaging, and medical codes. Use for PyHealth dataset loading, MIMIC-III/IV, eICU or OMOP prediction tasks, patient-level evaluation, mortality/readmission/length-of-stay modeling, medication recommendation, sleep staging, Trainer…
ml-training-recipes
Battle-tested PyTorch training recipes for all domains — LLMs, vision, diffusion, medical imaging, protein/drug discovery, spatial omics, genomics. Covers training loops, optimizer selection (AdamW, Muon), LR scheduling, mixed precision, debugging, and systematic experimentation. Use when training or fine-tuning…
nemo-mbridge-perf-expert-parallel-overlap
Validate and use MoE expert-parallel communication overlap in Megatron-Bridge, including overlapmoeexpertparallelcomm, delaywgradcompute, and flex dispatcher backends such as DeepEP and HybridEP.