Roadmap¶

This document outlines planned features and enhancements for shtym. Timing and scope may change based on feedback and contributions.

Near Term (High Priority)¶

Project-Local Profile Configuration¶

Support profile files in project directories (e.g., .shtym/profiles.toml), not just ~/.config/shtym/.

Why: Enable team-shared configurations committed to version control, with automatic profile discovery per project.

Example:

my-project/
├── .shtym/profiles.toml  # Team profiles
└── src/

cd my-project
stym run --profile team-standard pytest  # Automatically uses project profiles

Profile Inheritance¶

Allow profiles to extend other profiles to reduce duplication.

Example:

[profiles.base-summary]
system_prompt_template = "Summarize: $command"
user_prompt_template = "$stdout"

[profiles.fast-summary]
extends = "base-summary"
[profiles.fast-summary.llm_settings]
model_name = "gemma3:4b"

Why: Maintain related profiles more easily with shared configuration.

Environment Variable Expansion in Profiles¶

Enable ${VAR_NAME} syntax in profile configurations to separate secrets from configuration files.

Why: Keep API keys and credentials out of version control while sharing profile structure with teams.

Example:

[profiles.openai]
type = "llm"
[profiles.openai.llm_settings]
api_key = "${OPENAI_API_KEY}"
base_url = "${OPENAI_BASE_URL:-https://api.openai.com/v1}"

Structured Output Support¶

Enable LLM responses in structured formats (JSON) with schema validation.

Why: Make LLM outputs machine-readable for downstream processing, CI/CD integration, and programmatic analysis.

Example:

[profiles.test-analysis]
type = "llm"
system_prompt_template = "Analyze test results: $command"
user_prompt_template = "$stdout"
output_format = "json"
output_schema = """
{
  "type": "object",
  "properties": {
    "total_tests": {"type": "integer"},
    "passed": {"type": "integer"},
    "failed": {"type": "integer"},
    "errors": {
      "type": "array",
      "items": {"type": "string"}
    }
  }
}
"""

Use cases:

Parse test results for CI/CD metrics
Extract structured lint findings
Generate machine-readable summaries for automation

Template Variable Support¶

Enable user-defined variables in prompt templates with default values, allowing dynamic customization at runtime.

Why: Reuse the same profile with different parameters without creating multiple profiles. Useful for automation tools like nox.

Example:

[profiles.custom-summary]
type = "llm"
system_prompt_template = "Summarize from ${perspective:-general} perspective: $command"
user_prompt_template = "$stdout"

Usage:

# Use default value (general)
stym run --profile custom-summary pytest tests/

# Override with specific perspective
stym run --profile custom-summary --var perspective=security pytest tests/
stym run --profile custom-summary --var perspective=performance ruff check .

Long Context Handling¶

Support processing large command outputs that exceed LLM context limits through chunking and hierarchical summarization.

Why: Enable shtym to handle real-world scenarios with massive test suites, large diffs, or extensive logs without hitting token limits.

Example strategies:

Chunk large outputs and process separately
Hierarchical summarization (summarize chunks, then summarize summaries)
Extract critical sections (errors, failures) before processing
Stream processing for very large outputs

Use cases:

Large test suites with thousands of tests
Extensive log files
Large git diffs across many files
Build outputs from complex projects

Medium Term¶

OpenAI API Support¶

Add support for OpenAI's API as an LLM provider alongside Ollama.

Why: Provide flexibility to use different LLM providers based on availability, cost, or performance needs.

Azure Key Vault Integration¶

Fetch secrets from Azure Key Vault using akv:// URI scheme (via envresolve library).

Why: Enterprise-grade secret management with audit trails and automatic rotation support.

Example:

api_key = "akv://my-vault/openai-key"
# Or with variable expansion:
api_key = "akv://${VAULT_NAME}/openai-key"

Long Term (Lower Priority)¶

Command-Specific Prompt Presets¶

Provide built-in optimized prompts for common development commands (pytest, ruff, git, npm, etc.) with automatic detection.

Why: Make it easier to get started with effective prompts without manual prompt engineering.

Example:

stym run pytest tests/  # Automatically uses pytest-optimized prompt

External Prompt Import¶

Enable importing prompt configurations from URLs, files, or repositories.

Why: Share and reuse proven prompt configurations across teams and projects.

Example:

stym profile import https://example.com/prompts/pytest-detailed.toml

Additional LLM Providers¶

Extend support to Anthropic (Claude), Azure OpenAI Service, and other major LLM providers.

Why: Maximize flexibility and support diverse deployment scenarios.

Configuration Management Commands¶

Add stym config subcommands for validation and inspection:

stym config validate - Check profile syntax and configuration
stym config list - Show all available profiles
stym config show <profile> - Display profile details

Why: Improve debugging and discoverability of profiles.

Contributing¶

Interested in implementing a feature? Please open a GitHub Issue to discuss the approach before starting work.

Feedback¶

Have ideas for other features? Open an issue or start a discussion.