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.
git clone --depth 1 https://github.com/StamKavid/last-ds-mileWrote 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/commands/stamkavid/last-ds-mile/ds)<a href="https://agentmods.dev/commands/stamkavid/last-ds-mile/ds"><img src="https://agentmods.dev/badge/commands/stamkavid/last-ds-mile/ds.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 | $0.00012 | $0.00814 |
| Opus 5 | $0.00006 | $0.00407 |
| Sonnet 5 | $0.00002 | $0.00163 |
| Haiku 4.5 | $0.00001 | $0.00081 |
Grade A, and why
ds 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 6d 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.
What it actually says
Check .last-ds-mile/stages/ in the current project for existing stage output files
(00-frame.md through 12-deploy.md). Use the Glob tool on .last-ds-mile/stages/*.md
to see what already exists.
Print the pipeline map below, marking each stage done (✓) if its file exists, or next (→) for the first missing one in order:
0. /ds-frame Problem framing
1. /ds-data Data understanding
2. /ds-explore EDA
3. /ds-prep Cleaning + feature engineering <─┐
4. /ds-baseline Honest baseline │
5. /ds-validate Validation design │ loop back
6. /ds-model Modeling <─┤ (see /ds-iterate)
7. /ds-evaluate Evaluation + error analysis ───┘
7½. /ds-iterate Diagnose evaluate's findings, route back or proceed
8. /ds-explain Interpretation
9. /ds-report Communication
10. /ds-handoff Reproducibility & handoff
11. /ds-package Package + training/serving parity check
12. /ds-deploy Local endpoint + monitoring, drift, rollback
Stages 3–7 are not strictly linear: /ds-iterate reads /ds-evaluate's findings after
every pass and either routes back to /ds-prep, /ds-validate, or /ds-model with a
named diagnosis, or confirms the result is ready to proceed. Don't treat one pass
through 3–7 as automatically done — check whether .last-ds-mile/stages/07-iterate-log.md
exists and has a "proceed" verdict before recommending /ds-explain.
This command shows the map and stops. That is its whole job.
It is a status answer to "where do I stand", nothing more. If .last-ds-mile/stages/
doesn't exist yet, print the map, recommend /ds-frame as the first move, and explain
that each stage's output feeds the next. If stages exist, print the map and name the
first missing one.
Never route a task through this command. A request to build, evaluate, or ship a
model belongs to the data-science-project skill, which carries it to a result in the
same turn — reaching this map mid-task and stopping to ask which stage to start at is a
measured failure, not a hypothetical one: in a two-arm eval it cost two of three trials
their verdict entirely, while the unaided model simply answered. If you are carrying out
a task and find yourself here, you took a wrong turn: go back to data-science-project
and do the work.
When recommending the next stage: name the command matching the first missing stage in
order. If the user asks to skip ahead to /ds-model, /ds-report,
/ds-handoff, /ds-package, or /ds-deploy without the stages before it, apply
ds-method's Hard Gate handling: discipline gates (baseline, validation, slice
performance, pinned environment) get produced inline so the task still completes;
safety gates (/ds-package's parity check, /ds-deploy's monitoring/rollback) stop and
ask, since those are irreversible or externally visible. If /ds-evaluate exists but
/ds-iterate hasn't run yet, recommend /ds-iterate next, not /ds-explain.
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.
- 6d ago First seen · 56 lines · 12 tokens per session scan A f574f414e527
ds is a command published in the GitHub repository StamKavid/last-ds-mile (3 stars, last pushed 29d ago), licensed MIT. It adds 12 tokens to every session and 814 once invoked, about $0.0001 per session on Opus 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 2026-08-31.
Other commands, from other repositories
ab-test-design
You are a senior Data Science & AI/ML specialist. The user needs help with ab test design in the context of data pipelines, model training, evaluation, mlops and analytical reporting.
paper-trail-ingest
Ingest the citations of one SOTA into the registry. Extracts bibliographic sections, parses citations via the citation-parser sub-agent, resolves DOI via Crossref/S2, deduplicates against the registry, creates new candidate refs, and substitutes free-text citations with [[wikilinks]]. Required first step to make a…
paper-trail-new-paper
Start writing an academic paper (IMRaD structure) on a topic, with anti-hallucination citation verification at every step. Builds on existing audited SOTAs.
openehr-explain
One-stop router that explains or looks up any openEHR thing — auto-detects an archetype, a template, an RM/AM/BASE type, an RM structural concept, an ADL idiom, an AQL query or keyword, or a terminology code (replaces /archetype-explain, /template-explain, /type-spec, /rm-structure, /adl-idiom, /terminology).
nixtla-baseline-m4
Run baseline forecasting models on M4 Daily dataset.
create-scene
Create a single animated scene for mathematical visualization.