Roadmap¶

Keen Code is intentionally minimal. The goal is not to match every feature of every coding agent — it is to stay focused on what makes a terminal-first agent genuinely useful for day-to-day software engineering.

This roadmap is organized into three tiers. Tier 1 is what needs to exist before the tool is really trustworthy for serious work. Tier 2 expands capability meaningfully. Tier 3 is about depth and polish.

Items marked ✅ are shipped.

Tier 1 — Core Usability and Trust¶

✅ Pre-Approved Permissions¶

A mechanism to pre-approve tools for the current project. This is a way to trust the agent with certain tools without having to approve every single use or allow in-session approval.

✅ Web Fetch Tool¶

Let the agent fetch content from a URL and return it as text. HTML pages are converted to Markdown for readability. Network access will remain explicit and policy-controlled.

Tier 2 — Capability Expansion¶

✅ Native Code Review Mode¶

A first-class way to review changes rather than implement them. Point it at unstaged changes, a diff, or specific files and get findings ordered by severity — bugs, regressions, risks, missing tests. Optionally generate fix suggestions without applying them automatically.

✅ MCP Support¶

Add Model Context Protocol support so Keen can connect to external tools and systems through a standard layer. This is the extensibility path — instead of bundling one-off integrations, MCP lets users bring their own tools without waiting on core changes.

✅ Custom Commands and Reusable Workflows¶

Define reusable slash commands that package common prompts or workflow entry points. Commands can live in the repo (team-shared) or in user config (personal). Think /review, /release-notes, /fix-tests — the kind of repeated tasks every team has. Agent skills can be leveraged for this.

Persistent Memory¶

Retain useful context across sessions beyond static instruction files like AGENTS.md. Project-local and user-global memory, kept visible and editable so it never silently mutates behavior.

Subagents and Delegation¶

Let the agent delegate bounded subtasks — exploration, implementation, testing, review — to additional agents running in parallel. Keeps the main thread clean while handling larger tasks. All existing permission and filesystem boundaries still apply.

Tier 3 — Workflow Depth¶

Hooks and Automation¶

Hook points around tool activity for running formatters, validators, linters, or notifications automatically after edits. Keeps Keen Code fitting into established engineering workflows rather than working around them.

Export or share session transcripts for collaboration, debugging, or reproducibility. Local export first; remote sharing later.

Image Input¶

Attach local images or screenshots to prompts for models that support it. Most useful for frontend work and visual debugging tasks.

Out of Scope (for now)¶

Cloud task orchestration or remote agent execution
Complex auto-triggered workflows
Large collections of bundled niche tools
Provider-specific features that would create inconsistent behavior across providers

Contributing¶

If something on this list matters to you, issues and PRs are open. The issue templates include a provider/model request form for additions to providers/registry.yaml, and AGENTS.md has the short guide for adding a new tool.