provenance-reproducibility: what it makes your agent do, and what it refuses.

provenance-reproducibility answers the only question that ends the argument about a number: re-run it yourself. A content hash for the input, the commit that rebuilds the page, the period pinned to the record, an exit code a pipeline can read, and a written list of what a cold re-run could not re-judge.

The job provenance-reproducibility does

The test of a published figure is not whether it looks right, it is whether someone with no access to you and no reason to take your word can rebuild it and get the same bytes. This skill makes the provenance record a checkable sentence: this file, built by this engine, asked this question, produced these values, under these verdicts, at this exit code. It settles identity before computing, reproduces the original question rather than today's, insists the recorded commit actually rebuilds the page, and runs the reproduction cold on a machine that holds neither the prior cut nor your environment. Verdicts that depend on history cannot be re-judged cold, so the record names them instead of quietly dropping them.

What provenance-reproducibility refuses

A skill is worth what it stops as much as what it produces. These are the refusals written into provenance-reproducibility, taken from the rules and limits in its own file.

✓A record over bytes that can change. If the stored input can be overwritten, verify is theatre and the hash points at a file that no longer exists.
✓A verify that only runs where it was born. The reproduction imports nothing from the code under audit and runs the real entry point, not a re-implementation.
✓Silence about what cannot be re-judged. Unproven is a verdict and it is stated, because unproven and stated beats proven looking and false.
✓An exit code that lies. The contract is explicit, so a refused build cannot exit green and a pipeline can act on the result.

6 lines from the provenance-reproducibility skill file, verbatim

From skills/provenance-reproducibility/SKILL.md, unedited. Kymira skills are plain text you can read in full before you install them, which is the point of shipping a method rather than a binary.

The real test is a machine that has **neither the prior cut, nor the sibling
sources, nor your environment**. What reproduces there is what is actually
proven; everything else is proven only in the tree that wrote it. So: recompute
from the artifact's own bytes, import nothing from the code under audit, and run
the real entry point rather than a re-implementation of it. A verify that is
green only where it was born is not evidence.

The failure modes provenance-reproducibility addresses

These entries in the catalogue of 51 observed failure modes name this skill as what addresses them. Every one was seen in a real build or in our own audits, and none was invented.

The hash that disproves itselfThe verification is theater The stale fingerprintThe verification is theater The stack trace deliverableThe wrapper betrays the numbers

The tier provenance-reproducibility ships in

Tier	Price	What the tier holds	provenance-reproducibility
Free	$0	honest-dataviz and the eleven doctrine principles, published in full	Not in this tier
Core Pack	$199	the thirteen craft skills, one-time purchase, lifetime updates	Included
Everything	$349	all 22 skills, both libraries, the template bank, one-time purchase, lifetime updates	Included

Both paid tiers are a one-time purchase with lifetime updates, and all sales are final: the files are plain text you keep the moment they arrive, so judge the quality on the free skill and the published doctrine first. The pricing section states exactly what each tier contains.

Installing provenance-reproducibility

A skill is a folder with a SKILL.md file in it. Installing one is copying that folder where your agent looks for skills, with no build step and no dependencies.

mkdir -p ~/.claude/skills
cp -R Kymira/skills/provenance-reproducibility ~/.claude/skills/

Or open the unzipped pack in Claude Code and say install Kymira, which installs every skill in your tier, or double-click the installer for your system. Cursor reads the same skill as a rule file: copy provenance-reproducibility.mdc into .cursor/rules/. Skills are read at startup, so start a new session and ask your agent which skills it has available.

Not for

Reproduction proves the numeric path re-runs to the same values, not that the values are right. A wrong column or a wrong grain reproduces perfectly, forever, and a matching hash proves the input is the same input rather than a correct one. Definitions are where the largest class of confident wrong numbers lives, and this skill is silent on them.

Last updated 2026-08-09 · Dylan, founder · one of 22 packaged skills, each one a method written down, not a product tour.

See the tiers → All 22 skills The doctrine The failure catalogue