kornia-curation
Skill748 starsChanged 7 days agoApache-2.0View on GitHub
Use when a change touches a fact the skills, AGENTS.md or CONTRIBUTING.md describe (MSRV, publish set, pixi tasks, CI commands, features, crate layout), when you hit a repo trap worth recording, or when editing .claude/skills — keeps the guidance true and current. Trigger on "update the skills", "docs are stale", "add a skill", or any edit to Cargo.toml, pixi.toml, scripts/ or .github/workflows/.
We split this skill into the rules it gives an agent
3 rules. About 0 are common, 1 is distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 35
When rewording a restated fact, verify it against the source (e.g.
- Very specific: line 42
Reviews feed skills: Every kornia-code-review run ends with a "Skill gaps" section.
- Distinctive: line 47
Delete what is wrong: Remove or fix stale lines.
≈ “Common” is an estimate: at least 1 in 5 of the 308,232 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Keeping kornia-rs guidance current
Stale guidance is worse than none: agents follow it. Guidance points at the
source of truth; it restates a fact only when scripts/check_skills.py
cross-checks it.
Sources of truth
| Fact | Source | Restated in skills? |
| MSRV | root Cargo.toml rust-version | yes, in skills, AGENTS.md, CONTRIBUTING.md, Copilot instructions |
| Publish set / order | scripts/release_rust.sh (PUBLISH_ORDER, NOT_PUBLISHED) | no — link only; a sentence calling a PUBLISH_ORDER crate (kornia-x or short name) excluded / not published fails the check |
| pixi tasks | pixi.toml [tasks] / [feature.*.tasks] | yes, every pixi run [-e env] <task> is checked |
| CI commands | .github/workflows/*.yml | summarised; workflow file wins |
| Crates | crates/* | yes, names in kornia-arch are checked |
| Paths cited in skills | the tree | yes, backticked paths under crates/, scripts/, kornia-py/, kornia-cpp/, docs/, .github/ |
| Review eval cases | kornia-code-review/evals/ | yes, files present and patch.diff applies |
| Features | each crate's Cargo.toml | summarised |
| Release notes | changelog.d/ fragments; CHANGELOG.md only in release PRs | scripts/changelog.py check in CI |
| SIMD rules | crates/kornia-imgproc/SIMD.md | link only |
Rules
- Same-PR updates. A PR that changes a source above updates every skill
and doc that describes it.
pixi run skills-check(and the pre-commit hook, which also fires onCargo.toml,scripts/release_rust.shand workflow edits) catches the checked facts; the rest is on the author and reviewer. - Link, don't copy. New guidance names the file that owns the fact. If you
must restate a fact, add its check to
scripts/check_skills.pyin the same PR. When rewording a restated fact, verify it against the source (e.g.PUBLISH_ORDER), not only against the check: the check can be wrong too. - Record traps and findings. A non-obvious failure hit during a task (silent skip, wrong default, misleading error) becomes one line in the owning skill: the invariant and how to detect it. So does a reusable finding (an existing helper to call, a measured cost, a pattern that worked), with the file or command that shows it. No narratives, no dates.
- Reviews feed skills. Every
kornia-code-reviewrun ends with a "Skill gaps" section. Each gap becomes a one-line rule in the owning skill; each defect a review missed becomes a case in.claude/skills/kornia-code-review/evals/. A skill edit is done only when all eval cases still pass. - Delete what is wrong. Remove or fix stale lines; never add a contradicting line next to them.
- New skill: directory name == frontmatter
name, description starts with "Use " and says when to trigger, row added to the AGENTS.md table, andREQUIREDinscripts/check_skills.pyif contributors must read it. - Short. A skill is a checklist, not a tutorial. Cut before you add.
In this repository11 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- kornia-api-flavorSkill
- kornia-archSkill
- kornia-bindingsSkill
- kornia-code-reviewSkill
- kornia-coding-styleSkill
- kornia-simdSkill
- kornia-workflowSkill
- releaseSkill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.