Get This Tool
improv.sh
Pricing
- Model
- Free
Summary
Coding agents spend their first two turns asking you what you meant — burning tokens on discovery before writing a single line of code. improv cuts that loop by compiling your rough request into a structured spec before the agent ever sees it.
improv operates as a task harness: the @im call pulls repo context, detects your test commands, writes acceptance criteria, and packages shell validation steps into one spec the agent can implement on turn one. The loop infrastructure is the distinguishing piece — judges run your actual exit-code commands (npm test, typecheck, build), so done means your tests pass, not that the agent says it's done. The tool installs locally via curl with no external API keys required, and the Chrome extension brings the same engine into web-based chat interfaces. The 920-skill library and daily auto-research loop suggest the routing layer will keep growing — but the page offers no independent benchmarks to validate the token-savings figures cited.
Bottom line: Pick this for local agent workflows where you want structured specs and real shell validation closing the loop — it breaks down when your team needs shared task state across machines or an audit trail outside the .improv/tasks/ directory.
Community Performance Report Card
No community ratings yet. Be the first to rate this tool!
Community Benchmarks Community
Sign in to submit a benchmarkNo community benchmarks yet. Be the first to share a real-world data point.
Pros
Sign in to edit- Repo-aware spec compilation pulls your actual test commands and package scripts into the task before the agent starts, which means the agent implements against your real constraints instead of inventing them mid-run.
- Exit-code judges close the loop on real shell commands — npm test, typecheck, build — so you are not relying on the agent's self-assessment of whether it finished.
- Task memory persisted under .improv/tasks/ survives session boundaries, so an agent restarted mid-task picks up status and spec instead of starting the discovery cycle again.
- Local-first install with no external API keys required, which means the harness runs in air-gapped or locked-down environments where cloud tooling is blocked.
- Chrome extension and VS Code/Cursor Marketplace extension share the same local engine, so the spec compilation and judge loop work whether you are in the IDE or a browser-based chat interface — without switching tabs.
Cons
Sign in to edit- Task state is written to .improv/tasks/ on the local machine. Teams with more than one developer working the same codebase have no shared task state — there is no sync layer described on the page — so parallel agent runs on different machines produce divergent task records with no reconciliation path.
- The tool exposes no API surface, so teams that want to trigger improv from a CI pipeline or wrap it in a custom orchestration layer cannot. Teams hitting this wall move to harness frameworks that expose programmatic interfaces — at which point they are maintaining the prompt compilation logic themselves.
- The token-savings figures on the page (~613 tokens median) are vendor-reported with no independent reproduction methodology described. Teams making adoption decisions based on cost reduction should treat these numbers as illustrative until they run their own baseline comparison.
- Chrome extension installation requires either the Chrome Web Store or a manual sideload script — neither path is available in Firefox or Safari. Teams on non-Chromium browsers are limited to the terminal install, losing the browser chat integration entirely.
Community Reviews
Sign in to write a reviewNo reviews yet. Be the first to share your experience.
About
- Platforms
- VS Code, Cursor, Claude Code, terminal, Chrome
- API Available
- No
- Self-Hosted
- Yes
- Last Updated
- 2026-07-22T00:23:42.178Z
Best For
Who it's for
- Developers using AI coding agents in IDEs or terminal
- Workflows requiring structured task specs and shell validation
- Local setups needing no external API keys
What it does well
- Compiling vague coding requests into actionable specs for agents
- Running iterative loops with real test/judge commands until completion
- Enhancing prompts in VS Code, Cursor, or terminal without leaving the editor
Integrations
Discussion Community
Sign in to commentNo discussion yet. Sign in to start the conversation.
Spotted incorrect or missing data? Join our community of contributors.
Sign Up to ContributeCommunity Notes & Tips Community
Sign in to contributeBe the first to contribute. General notes, observations, gotchas, and tips from people who use this tool day-to-day.
Frequently Asked Questions
- Is improv.sh free?
- Yes — improv.sh is fully free to use. There is no paid tier.
- Is improv.sh open source?
- Yes. improv.sh is open source.
- Can I self-host improv.sh?
- Yes. improv.sh supports self-hosting on your own infrastructure.
- What platforms does improv.sh support?
- improv.sh is available on: VS Code, Cursor, Claude Code, terminal, Chrome.
Hours Saved & ROI Stories Community
Sign in to contributeBe the first to contribute. Concrete time/cost savings, with context. e.g. "Cut my code review backlog from 4h to 45m per week."
Curated lists that include this category
Most AI coding agents rediscover the task on every run, interviewing you across multiple turns before producing anything useful. improv addresses this with a task harness: a single @im or /im call compiles a vague request into a structured spec containing objective, context, constraints, deliverables, acceptance criteria, and shell validation commands — then passes that spec to the agent so implementation starts on turn one. The loop closes not with LLM self-report but with real exit-code judges; the agent implements, the judges run, failures feed back in, and the cycle repeats until exit code 0.
The differentiating feature is the loop infrastructure. The vendor describes slice-by-slice execution checklists — inspect, implement, verify, report — combined with task memory persisted under .improv/tasks/. Agents forget context between sessions; the task lifecycle ID and status tracking mean the harness does not. Repo-aware context detection pulls package scripts, test commands, and agent instruction files from the workspace automatically, so the spec reflects your actual stack rather than a generic template.
improv installs locally via a curl script with no API key required, and the Chrome extension sideloads the same engine into browser-based chat interfaces including Claude Code, Codex, and Cowork. VS Code and Cursor users get a Marketplace extension that lets them diff, apply, and run judges without leaving the editor. The 920-skill library grows via a daily self-improving search-score-adopt pipeline, which the vendor states runs automatically. There is no API surface documented on the page, so integrating improv into a custom CI pipeline or programmatic workflow is not supported by the current architecture.