Borrowing it
Nothing to install: this file belongs to radiantlogicinc/fastworkflow. 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/radiantlogicinc/fastworkflow/main/.claude/skills/fastworkflow-nlu-pipeline-reference/SKILL.mdgit clone --depth 1 https://github.com/radiantlogicinc/fastworkflowWrote 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/radiantlogicinc/fastworkflow/fastworkflow-nlu-pipeline-reference)<a href="https://agentmods.dev/skills/radiantlogicinc/fastworkflow/fastworkflow-nlu-pipeline-reference"><img src="https://agentmods.dev/badge/skills/radiantlogicinc/fastworkflow/fastworkflow-nlu-pipeline-reference.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00300 | $0.09344 |
| Opus 5 | $0.00150 | $0.04672 |
| Sonnet 5 | $0.00060 | $0.01869 |
| Haiku 4.5 | $0.00030 | $0.00934 |
Grade C, and why
fastworkflow-nlu-pipeline-reference scanned grade C with 1 finding 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 8d 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.
Recursive force deletehighDestructive command
rm -rf with a variable or a broad path is one typo away from removing the wrong tree.
in-process, or `rm -rf ~/.dspy_cache/ ./.dspy_cache/` (dir verified via How it starts
The opening of the file, as written. The whole thing — 506 lines — stays where its author put it; the contents beside it link to each section on GitHub.
fastWorkflow NLU Pipeline Reference
Everything below is verified against source at v2.22.2 (commit c33b9a5), 2026-07-09,
except the passages marked "2026-08-02", which were re-verified against the working
tree after wave 1 of epic fix-551 (held-out evaluation, determinism, artifact
versioning, and the reserved-label split) landed. File paths are repo-relative. Trust
this document over CLAUDE.md and README where they disagree — known doc rot is listed at
the end.
2026-08-02 — the biggest change since this skill was written.
wildcardis no longer a catch-all.fastworkflow/nlu_labels.pysplits it into two reserved labels with different meanings and different training rules; §2 has the detail. Anything you remember about "the wildcard class" is at best half right now.
fastworkflow/model_pipeline_training.pyis under active edit as R1 is wired in, so the line numbers below drift. Prefer the symbol names;grep -nfor them.
When to use / when NOT to use
| You need... | Use |
|---|---|
| How intent detection / parameter extraction actually work here; what a threshold means; where a magic number lives | This skill |
| A symptom-to-fix triage table for a live failure | fastworkflow-debugging-playbook |
| Every env var with defaults and consumers | fastworkflow-config-and-flags |
| tau-bench / tau2-bench harness, pass^k, simulator mechanics | fastworkflow-taubench-reference |
| Why the architecture is shaped this way; invariants | fastworkflow-architecture-contract |
| Running train/build/CLI commands operationally | fastworkflow-run-and-operate |
| Measuring model quality instead of eyeballing | fastworkflow-diagnostics-and-tooling |
| Growing seeds/personas and judging whether run B beat run A | fastworkflow-intent-training-convergence |
| Variance/pass^k math, calibration analysis recipes | fastworkflow-proof-and-analysis-toolkit |
Glossary (one line each, used throughout)
| Term | Meaning here |
|---|---|
| fine-tuning | Continuing training of a pretrained model's weights on your small labeled dataset |
| logits | Raw per-class scores from the classifier head, before normalization |
| softmax confidence | max(softmax(logits)) — the top class's probability; this repo's only "confidence" signal |
| calibration | How well confidence tracks actual correctness. This repo tunes decision thresholds on held-out data but never calibrates the probabilities themselves (no temperature scaling) |
| embedding | Fixed-length vector representing text; here the DistilBERT [CLS] token's last hidden state (cache_matching.py:51) |
| cosine similarity | Angle-based vector similarity in [~-1, 1]; 1.0 = same direction |
| Levenshtein distance | Minimum single-character edits between strings; "normalized" = divided by the longer length, so 0.0 = identical |
| LabelEncoder | sklearn utility mapping label strings ↔ integer class ids (the ONLY load-bearing sklearn use in intent detection) |
| weighted F1 | Per-class harmonic mean of precision/recall, averaged weighted by class frequency |
| NDCG@3 | Ranking score: 1.0 if the true label is ranked 1st, discounted by 1/log2(rank+1) if 2nd/3rd, 0 if absent from top-3 |
| DSPy | Framework that compiles typed "Signatures" (input/output field specs) into LLM prompts via modules like ChainOfThought |
| ChainOfThought | DSPy module that makes the LLM emit reasoning before the output fields |
| LabeledFewShot | DSPy optimizer that stuffs labeled examples into the prompt (few-shot = examples-in-prompt) |
| litellm | Client library exposing one API over many LLM providers via provider/model strings |
| NOT_FOUND | String sentinel (env var NOT_FOUND, value "NOT_FOUND") marking an unextracted parameter |
What ships with it
4 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.
- 8d ago First seen · 506 lines · 300 tokens per session scan C ca8e9cd0cd5f
fastworkflow-nlu-pipeline-reference is a skill published in the GitHub repository radiantlogicinc/fastworkflow (52 stars, last pushed 4d ago), licensed Apache-2.0. It adds 300 tokens to every session and 9,344 once invoked, about $0.0015 per session on Opus 5. A static security scan graded it C with 1 finding (recursive force delete). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
agent-platform-rag-engine-management
Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…
agent-platform-model-registry
Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.
foundry-config-setup
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded projectendpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.
google-cloud-solution-agentic-analytics-spark-knowledge-catalog
Discovers requirements and generates guidance to design and deploy a governed, secure agentic-analytics solution for data that's distributed across Google Cloud, other cloud providers, or on-premises. Data that's outside Google Cloud (such as data from Databricks, Snowflake, Salesforce, SAP, or Oracle systems) is…
training-check
Interactively monitor training metrics from the current Codex session, periodically checking WandB or fallback logs for NaN, divergence, plateaus, and broken runs.
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.