diff --git a/.agents/skills/schema-analysis/SKILL.md b/.agents/skills/schema-analysis/SKILL.md
new file mode 100644
index 0000000..eb84499
--- /dev/null
+++ b/.agents/skills/schema-analysis/SKILL.md
@@ -0,0 +1,145 @@
+---
+name: schema-analysis
+description: >
+  Infer schemas and generate types (Rust, TypeScript, Kotlin, JSON Schema) from
+  data files in JSON, YAML, XML, TOML, CBOR, or BSON. Use when you need to
+  understand the structure of a data file, generate type definitions for code,
+  produce a JSON Schema, or compare the shapes of multiple data files.
+license: MIT
+compatibility: Requires npx (Node.js 18+), or uvx (Python 3.8+), or cargo
+metadata:
+  author: QuartzLibrary
+  version: "0.7.0"
+---
+
+# schema_analysis
+
+Infer schemas from data files and generate typed code or JSON Schema.
+
+## Running
+
+```bash
+npx schema_analysis [OPTIONS] [FILES]...
+# or
+uvx schema_analysis [OPTIONS] [FILES]...
+```
+
+Input format is auto-detected from file extensions (`.json`, `.yaml`/`.yml`, `.xml`, `.toml`, `.cbor`, `.bson`, use `--format` to override).
+Reads from stdin if no files are given (defaults to JSON; use `--format` to override).
+
+## Options
+
+| Flag | Description |
+|---|---|
+| `--format <FORMAT>` | Override input format: `json`, `yaml`, `xml`, `toml`, `cbor`, `bson` |
+| `--output <OUTPUT>` | Output mode (default: `schema`) |
+| `--name <NAME>` | Root type name for code generation (default: `Root`) |
+| `--compact` | Compact JSON output. **Always use this flag.** |
+| `--minimal` | Omit statistics (counts, samples, min/max). Structure only. |
+
+### Output modes (`--output`)
+
+| Value | Produces |
+|---|---|
+| `schema` | Inferred schema JSON with field metadata |
+| `rust` | Rust structs with `Serialize`/`Deserialize` derives |
+| `typescript` | TypeScript `interface` definitions |
+| `typescript-alias` | TypeScript `type` alias definitions |
+| `kotlin` | Kotlin data classes (Jackson) |
+| `kotlin-kotlinx` | Kotlin data classes (Kotlinx Serialization) |
+| `json-schema` | JSON Schema (Draft 2020-12) |
+| `shape` | Abstract shape representation |
+
+## Choosing the right invocation
+
+**"I need to understand the shape of this data"** - use the default schema output:
+
+```bash
+npx schema_analysis data.json --compact --minimal
+```
+
+This gives you the type tree with nullability/optionality flags and nothing else.
+
+**"I need types for code I'm writing"** - generate them directly:
+
+```bash
+npx schema_analysis data.json --output typescript --compact --name ApiResponse
+npx schema_analysis data.json --output rust --compact --name MyData
+npx schema_analysis data.json --output kotlin --compact --name Event
+```
+
+Paste the output straight into the target file.
+
+**"I need a JSON Schema for validation or documentation":**
+
+```bash
+npx schema_analysis data.json --output json-schema --compact
+```
+
+**"I have data in memory, not in a file"** - pipe it through stdin:
+
+```bash
+echo '{"key":"value"}' | npx schema_analysis --format json --compact --minimal
+```
+
+Always pass `--format` when reading from stdin.
+
+**"I have multiple sample files and need the unified schema":**
+
+```bash
+npx schema_analysis sample1.json sample2.json sample3.json --compact --minimal
+```
+
+Fields present in some files but not others will have `"may_be_missing": true`.
+
+## Reading the schema output
+
+The schema is a recursive JSON tree. Each node has a `type` (`"struct"`, `"sequence"`, `"string"`, `"integer"`, `"float"`, `"boolean"`, `"null"`).
+
+Struct nodes have `"fields"`: a map of field name to field descriptor.
+Sequence nodes have `"field"`: the schema of the array elements.
+
+Every field descriptor carries four flags:
+
+| Flag | Meaning |
+|---|---|
+| `may_be_null` | Was `null` in at least one record |
+| `may_be_normal` | Had a non-null value in at least one record |
+| `may_be_missing` | Was absent from at least one record (shows optionality across files) |
+| `may_be_duplicate` | Key appeared more than once in the same record (relevant for XML) |
+
+When `--minimal` is **not** used, extra statistics appear: `count`, `samples`, `min`/`max` (numbers), `min_max_length` (strings), `trues`/`falses` (booleans). Use the full output only when you need those statistics.
+
+### Example
+
+Input: `{"name":"Alice","age":30,"scores":[95,87]}`
+
+`--compact --minimal` output:
+
+```json
+{"type":"struct","fields":{"name":{"may_be_null":false,"may_be_normal":true,"may_be_missing":false,"may_be_duplicate":false,"type":"string"},"age":{"may_be_null":false,"may_be_normal":true,"may_be_missing":false,"may_be_duplicate":false,"type":"integer"},"scores":{"may_be_null":false,"may_be_normal":true,"may_be_missing":false,"may_be_duplicate":false,"type":"sequence","field":{"may_be_null":false,"may_be_normal":true,"may_be_missing":false,"may_be_duplicate":false,"type":"integer"},"context":null}},"context":null}
+```
+
+`--output typescript` output:
+
+```typescript
+export interface Root {
+    name: string;
+    age: number;
+    scores: number[];
+}
+```
+
+`--output rust --name MyData` output:
+
+```rust
+use serde_derive::Deserialize;
+use serde_derive::Serialize;
+
+#[derive(Default, Debug, Clone, PartialEq, Serialize, Deserialize)]
+pub struct MyData {
+    pub name: String,
+    pub age: i64,
+    pub scores: Vec<i64>,
+}
+```