Sculptorvsharness
Generated from the two spec rows. Green marks the better value where a spec has a clear direction. Everything else is just different.
Sculptor
Imbue · Agent harness
OSS
Panel
5.63 spec wins
- Reliability
- 5.2
- Usefulness
- 5.8
- Cost
- 6.8
- Longevity
- 4.7
“It ships a feature called CI Babysitter, which is the most accurate name anyone in this category has managed.”
harness
wedow · Agent harness
OSS
Panel
5.61 spec wins
- Reliability
- 5.0
- Usefulness
- 5.5
- Cost
- 7.3
- Longevity
- 4.7
“It can sign in with your ChatGPT or Claude subscription, which is a large decision for a hundred lines of shell script to be making.”
Spec by spec
| Spec | Sculptor | harness |
|---|---|---|
| Architecture | ||
| Category | Agent harness | Agent harness |
| Runs | local | local |
| Platforms | macos, linux | macos, linux |
| Context window | not documented | not documented |
| Protocols | ||
| MCP client | No | No |
| MCP server | No | No |
| Capabilities | ||
| Runs terminal commands | Yes | Yes |
| Multi-file edits | Yes | Yes |
| Git operations | Yes | No |
| Browser control | No | No |
| Sandboxed execution | NoThe marketing page says every agent runs in its own container, but the help docs state the default workspace is a git worktree and describe a Docker container backend as experimental. | No |
| Multi-agent orchestration | Yes | No |
| Headless / CI mode | NoThere is no headless mode, but a CI Babysitter watches open pull requests and dispatches an agent to fix failing checks and merge conflicts. | Yes |
| Models | ||
| Backbone | Claude | Anthropic, OpenAI, Groq, ChatGPT subscription, Claude subscription |
| Bring your own model | Yes | Yes |
| Local models | No | No |
| Cost | ||
| Pricing model | byok | byok |
| Starts at | n/a | $0/mo |
| Free tier | Yes | Yes |
| Bring your own key | Yes | Yes |
| Openness | ||
| Open source | Yes | Yes |
| License | MIT | MIT |
| GitHub stars | 234 | 169 |
Which one would each critic pick
| Critic | Sculptor | harness | Pick |
|---|---|---|---|
| El Juez | — | — | not enough reviews |
| El Amigo | 6.3 | 6.0 | Sculptor — Pick it if you want three agents attempting a task on three branches and a pull request at the end; pick Claude Squad if you would rather that happened in a terminal. |
| El Crítico | 5.3 | 4.5 | Sculptor — The product page says every agent runs in its own container; the help documentation says the default workspace is a git worktree and the container backend is experimental. |
| El Profesor | 6.5 | 6.3 | Sculptor — It drives the wrapped agent as a streaming-JSON process with the control protocol enabled and substitutes its own ask-user and plan tools, which is interception done properly. |
| La Inversora | 5.3 | 5.5 | harness — A hundred and sixty-seven stars against more than a thousand forum mentions in a year: this is the most-discussed and least-installed row I have read today. |
| La Jefa | 4.8 | 4.3 | Sculptor — It creates pull requests and dispatches an agent at failing checks, and I cannot tell an auditor who authorised any of it, on a platform list that excludes Windows. |
| El Hacker | 5.8 | 7.3 | harness — MIT, Homebrew or the AUR, and the entire dependency list is bash, jq and curl, with plugins in whatever language I feel like writing them in. |
Picks are derived from each critic's own scores. Humans vote on matchups on the duels page.