trust-the-upload: what it makes your agent do, and what it refuses.

trust-the-upload holds the five invariants that protect a user's confidence that nothing was lost, guessed or silently changed. Raw bytes are immutable, state commits atomically, an unreadable state is never overwritten, ambiguity becomes a question, and identity comes from what a file is, never from when it arrived.

The job trust-the-upload does

A reporting tool's real product is trust, and trust is lost in the plumbing rather than in the charts. This skill is the set of rules that protect the upload path, each one written with the measured data loss bug that made it law: an archive under names that are never reused, a single commit function so no failure path can update history without config, a staging directory promoted by rename only on success, and a refusal that names the file, the damage avoided and the fix in the user's own words. It also carries the sensitivity boundary, which decides what never reaches a page at all, and a triage runbook for the morning a green check goes red.

What trust-the-upload refuses

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

✓Reusing a name. A deleted file does not free its name, or the history quietly points at different bytes while provenance still reads verified.
✓A half written state. Config, history and published output commit together or not at all, and a refused upload never replaces a spec.
✓Writing state it could not read. A system that cannot parse its own history refuses rather than rebuilding it from an assumption.
✓A refusal with no way forward. Every refusal names the file, what would have been damaged, that nothing changed, and the specific fix.
✓Raw data on a page. Source exports stay in the landing zone, sensitive columns are named to the operator before the first chart, and a page for a wider audience shows groups rather than rows.

10 lines from the trust-the-upload skill file, verbatim

From skills/trust-the-upload/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.

- **Raw files never leave the landing zone.** `data/raw/` is the only
  home for source exports. Never copy raw data into a report directory,
  a chat message, or an example when a file path will do; work from the
  path and let code read the bytes.
- **Flag sensitive columns before the first chart.** On first read of any
  new source, name the columns that identify people or expose terms:
  emails, names, addresses, compensation, individual customer pricing,
  account numbers. Report the list to the operator and ask which may
  appear on a page BEFORE building anything that displays them.
- **Aggregate before you share.** A page meant for an audience wider than

The failure modes trust-the-upload 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 quantity slip the price screen missesThe number is plausible, and wrong The encoding that crashes the pipelineThe file lies The substring that misfiles a customerThe file lies The screen calibrated by its own targetThe verification is theater The honest refusal that catches nothingThe verification is theater The rerun that trampled the good packThe wrapper betrays the numbers

The tier trust-the-upload ships in

Tier	Price	What the tier holds	trust-the-upload
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 trust-the-upload

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/trust-the-upload ~/.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 trust-the-upload.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

It governs the path a file travels, not the arithmetic at the end of it. A workspace that never loses a byte can still publish a wrong number, and this skill will not notice. It also cannot recover data a system lost before these invariants were in place.

Last updated 2026-08-29 · 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