feat: add text input support

AGENTS.md

@@ -120,6 +120,7 @@ High-level approach.
 Available specs:
 
 - `llm-provider.md` - LLM provider abstraction, OpenAI integration, design choices
+- `text-input.md` - Direct text input via --text option, alternative to file input
 
 Specification format: Abstract and Requirements sections.
 
@@ -135,6 +136,7 @@ Available test cases:
 - `image_multimodal.md` - Image input for multimodal prompts
 - `image_generate.md` - Image generation and editing command
 - `error_handling.md` - Error scenarios and messages
+- `text_input.md` - Direct text input via --text option
 
 ### Test case template
 

README.md

@@ -8,7 +8,7 @@ Idea is simple, imagine you need to generate some docs using LLM as part of CI,
 
 > [!TIP]
 > This README was generated with trickery
-> trickery generate -i ./prompts/trickery_readme.md > README.md
+> trickery generate ./prompts/trickery_readme.md > README.md
 
 
 ## Demo
@@ -28,7 +28,17 @@ trickery --help
 
 ```sh
 export OPENAI_API_KEY=s....d
-trickery generate -i ./prompts/trickery_readme.md > README.md
+trickery generate ./prompts/trickery_readme.md > README.md
+```
+
+### Using with OpenAI-compatible gateways
+
+You can use trickery with any OpenAI-compatible API gateway (like LiteLLM, Azure OpenAI, or local models) by setting the `OPENAI_BASE_URL` environment variable:
+
+```sh
+export OPENAI_API_KEY=your-key
+export OPENAI_BASE_URL=http://localhost:4000/v1
+trickery generate ./prompts/my_prompt.md
 ```
 
 Input file could be any text file, with Jinja2-like template variables, like `{{"{{app_version}}"}}`. To set this variables, please use `-v` flag, like `-v app_version=1.0.0`.

docs/image-generation.md

@@ -4,9 +4,9 @@ Trickery supports generating and editing images using OpenAI's Responses API wit
 
 ## CLI Arguments
 
-### `--input <PATH>` / `-i <PATH>`
+### `[INPUT]` (positional) or `-i <INPUT>`
 
-Path to the prompt template file. Supports Jinja2-style `{{ variable }}` substitution.
+Input prompt: file path or direct text (auto-detected). Supports Jinja2-style `{{ variable }}` substitution when using a file.
 
 ### `--save <PATH>` / `-s <PATH>` (optional)
 
@@ -80,13 +80,13 @@ Template variables for prompt substitution.
 
 ```bash
 # Simple generation with explicit filename
-trickery image -i prompts/generate_diagram.md --save docs/images/colorful-architecture.png
+trickery image prompts/generate_diagram.md --save docs/images/colorful-architecture.png
 
 # Auto-generated filename (e.g., generate_diagram-a3f5x.png)
-trickery image -i prompts/generate_diagram.md
+trickery image prompts/generate_diagram.md
 
 # With quality settings
-trickery image -i prompts/generate_diagram.md -s architecture.png \
+trickery image prompts/generate_diagram.md -s architecture.png \
   --size 1536x1024 \
   --quality high
 ```
@@ -101,12 +101,12 @@ See [prompts/generate_diagram.md](../prompts/generate_diagram.md) for the prompt
 
 ```bash
 # Make an image look realistic
-trickery image -i prompts/make_realistic.md \
+trickery image prompts/make_realistic.md \
   --image test_data/example_images/image1.png \
   --save output.png
 
 # Edit with custom instruction
-trickery image -i prompts/edit_image.md \
+trickery image prompts/edit_image.md \
   --image test_data/example_images/image2.png \
   --save modified.png \
   -v instruction="make it green on pink"
@@ -117,7 +117,7 @@ See [prompts/make_realistic.md](../prompts/make_realistic.md) and [prompts/edit_
 ### Highlight Areas in Image
 
 ```bash
-trickery image -i prompts/highlight_humans.md \
+trickery image prompts/highlight_humans.md \
   --image test_data/example_images/image3.jpg \
   --save highlighted.png
 ```
@@ -127,7 +127,7 @@ See [prompts/highlight_humans.md](../prompts/highlight_humans.md) for the prompt
 ### With Template Variables
 
 ```bash
-trickery image -i prompts/generate_icon.md \
+trickery image prompts/generate_icon.md \
   --save icon.png \
   -v subject="rocket" \
   -v style="flat design"
@@ -139,7 +139,7 @@ See [prompts/generate_icon.md](../prompts/generate_icon.md) for the prompt templ
 
 ```bash
 # Combine elements from multiple images
-trickery image -i prompts/edit_image.md \
+trickery image prompts/edit_image.md \
   --image test_data/example_images/image1.png \
   --image test_data/example_images/image2.png \
   --save composite.png \
@@ -149,7 +149,7 @@ trickery image -i prompts/edit_image.md \
 ### Transparent Background
 
 ```bash
-trickery image -i prompts/generate_icon.md \
+trickery image prompts/generate_icon.md \
   --save logo.png \
   --background transparent \
   --format png \
@@ -160,7 +160,7 @@ trickery image -i prompts/generate_icon.md \
 ### JSON Output
 
 ```bash
-trickery image -i prompts/generate_diagram.md --save result.png -o json
+trickery image prompts/generate_diagram.md --save result.png -o json
 ```
 
 Output:

docs/input-images.md

@@ -34,7 +34,7 @@ Unknown extensions default to PNG MIME type.
 ### Single Local Image
 
 ```bash
-trickery generate -i prompts/describe_image.md --image test_data/example_images/image2.png
+trickery generate prompts/describe_image.md --image test_data/example_images/image2.png
 ```
 
 Where `prompts/describe_image.md` contains:
@@ -45,13 +45,13 @@ Describe what you see in this image in detail.
 ### Image from URL
 
 ```bash
-trickery generate -i prompts/describe_image.md --image https://www.google.com/images/branding/googlelogo/2x/googlelogo_color_272x92dp.png
+trickery generate prompts/describe_image.md --image https://www.google.com/images/branding/googlelogo/2x/googlelogo_color_272x92dp.png
 ```
 
 ### Multiple Images
 
 ```bash
-trickery generate -i prompts/catalog_images.md \
+trickery generate prompts/catalog_images.md \
   --image test_data/example_images/image1.png \
   --image test_data/example_images/image2.png \
   --image test_data/example_images/image3.jpg
@@ -61,13 +61,13 @@ trickery generate -i prompts/catalog_images.md \
 
 ```bash
 # Low detail for quick classification
-trickery generate -i prompts/describe_image.md --image test_data/example_images/image1.png --image-detail low
+trickery generate prompts/describe_image.md --image test_data/example_images/image1.png --image-detail low
 ```
 
 ### Combined with Variables
 
 ```bash
-trickery generate -i prompts/review_ui.md \
+trickery generate prompts/review_ui.md \
   --image test_data/example_images/image2.png \
   --var focus="accessibility" \
   --var format="bullet points"
@@ -100,7 +100,7 @@ Image support requires a vision-capable model. Recommended models:
 
 Example with explicit model:
 ```bash
-trickery generate -i prompts/describe_image.md --image test_data/example_images/image1.png --model gpt-5.2
+trickery generate prompts/describe_image.md --image test_data/example_images/image1.png --model gpt-5.2
 ```
 
 ## Token Considerations

prompts/trickery_readme.md

@@ -19,7 +19,7 @@ Idea is simple, imagine you need to generate some docs using LLM as part of CI,
 
 > [!TIP]
 > This README was generated with trickery
-> trickery generate -i ./prompts/trickery_readme.md > README.md
+> trickery generate ./prompts/trickery_readme.md > README.md
 
 
 ## Demo
@@ -39,7 +39,7 @@ trickery --help
 
 ```sh
 export OPENAI_API_KEY=s....d
-trickery generate -i ./prompts/trickery_readme.md > README.md
+trickery generate ./prompts/trickery_readme.md > README.md
 ```
 
 Input file could be any text file, with Jinja2-like template variables, like `{{"{{app_version}}"}}`. To set this variables, please use `-v` flag, like `-v app_version=1.0.0`.

specs/text-input.md

@@ -0,0 +1,80 @@
+# Text Input
+
+## Abstract
+
+Trickery's input supports both file paths and direct text, with auto-detection. If the provided value exists as a file, it reads from the file; otherwise, it treats the value as direct prompt text. Input is typically provided as a positional argument.
+
+## Requirements
+
+### Input Methods
+
+Input is provided as a positional argument:
+
+```bash
+trickery generate "prompt text"
+trickery generate prompts/greeting.md
+```
+
+The `-i` flag is also supported for backwards compatibility but positional is preferred.
+
+### Input Auto-Detection
+
+Once input is provided (either way), this logic applies:
+1. Check if the input value exists as a file on disk
+2. If file exists: read content from the file
+3. If file doesn't exist: use the input value directly as prompt text
+
+### Behavior
+
+```bash
+# File input (file exists, content read from file)
+trickery generate prompts/greeting.md
+
+# Text input (not a file, used as direct prompt)
+trickery generate "Write a haiku"
+```
+
+- Template variables work with both: `--var name=Alice`
+- For `image` command, output filename defaults to `image-xxxxx.png` when input is text
+
+### Long Text Support
+
+Positional input supports:
+
+- Multi-line strings (using shell quoting)
+- Special characters and Unicode
+- Very long prompts (limited only by shell argument length)
+
+### Shell Integration
+
+Examples of passing long text:
+
+```bash
+# Multi-line with shell quoting
+trickery generate "Line 1
+Line 2
+Line 3"
+
+# Using heredoc
+trickery generate "$(cat <<'EOF'
+You are a helpful assistant.
+
+Please analyze the following:
+- Point 1
+- Point 2
+EOF
+)"
+```
+
+## Design Choices
+
+### Why keep -i as fallback?
+
+1. Backwards compatibility with older scripts
+2. Useful when input looks like a flag (edge case)
+
+### Why auto-detect instead of separate options?
+
+1. Simpler API: one input concept
+2. Intuitive behavior: file paths look like file paths, text looks like text
+3. No ambiguity in practice: prompts rarely look like existing file paths

src/commands/generate.rs

@@ -1,6 +1,6 @@
 use clap::{Args, ValueHint};
 use serde::{Deserialize, Serialize};
-use std::path::PathBuf;
+use std::path::Path;
 use tokio::fs::read_to_string;
 
 use super::super::trickery::generate::{generate_from_template, GenerateConfig};
@@ -31,14 +31,22 @@ fn parse_key_val(s: &str) -> Result<(String, Value), String> {
 }
 
 #[derive(Args)]
+#[command(
+    args_conflicts_with_subcommands = true,
+    override_usage = "trickery generate [INPUT] [OPTIONS]"
+)]
 pub struct GenerateArgs {
-    /// Path to the input prompt file
-    #[arg(short, long, value_hint = ValueHint::FilePath)]
-    input: Option<PathBuf>,
+    /// Input prompt: file path or direct text (auto-detected)
+    #[arg(index = 1, value_name = "INPUT", value_hint = ValueHint::FilePath)]
+    pub input_positional: Option<String>,
+
+    /// Input prompt: file path or direct text (auto-detected)
+    #[arg(short, long = "input", value_name = "INPUT", value_hint = ValueHint::FilePath)]
+    pub input_option: Option<String>,
 
     /// Variables to be used in prompt
     #[arg(short, long="var", value_parser = parse_key_val, number_of_values = 1)]
-    vars: Vec<(String, Value)>,
+    pub vars: Vec<(String, Value)>,
 
     /// Model to use (e.g., gpt-5.2, gpt-5-mini, o1, o3-mini)
     #[arg(short, long)]
@@ -65,30 +73,45 @@ fn parse_reasoning_level(s: &str) -> Result<ReasoningLevel, String> {
     s.parse()
 }
 
+/// Resolve input to template content.
+/// If input exists as a file, read from file; otherwise treat as direct text.
+async fn resolve_input(input: &str) -> Result<String, Box<dyn std::error::Error>> {
+    let path = Path::new(input);
+    if path.exists() {
+        read_to_string(path)
+            .await
+            .map_err(|e| format!("Failed to read input file '{}': {}", path.display(), e).into())
+    } else {
+        Ok(input.to_string())
+    }
+}
+
+impl GenerateArgs {
+    /// Get input from either positional or -i option
+    pub fn get_input(&self) -> Option<&String> {
+        self.input_positional
+            .as_ref()
+            .or(self.input_option.as_ref())
+    }
+}
+
 impl CommandExec<GenerateResult> for GenerateArgs {
     async fn exec(
         &self,
         context: &impl super::CommandExecutionContext,
     ) -> Result<Box<dyn CommandResult<GenerateResult>>, Box<dyn std::error::Error>> {
-        let input_path = match &self.input {
-            Some(path) => path,
-            None => return Err("Input file path is required".into()),
-        };
+        let input = self
+            .get_input()
+            .ok_or("Input required: use positional arg or -i (file path or text)")?;
+
+        let template = resolve_input(input).await?;
 
         let input_variables: HashMap<String, Value> = self
             .vars
             .iter()
             .map(|(k, v)| (k.clone(), v.clone()))
             .collect();
 
-        let template: String = read_to_string(input_path).await.map_err(|e| {
-            format!(
-                "Failed to read input file '{}': {}",
-                input_path.display(),
-                e
-            )
-        })?;
-
         let images: Vec<String> = self.image.clone();
 
         let config = GenerateConfig {