agentsmith
Universal, model-agnostic operating harness for AI agents (Claude, Codex, Gemini, …) — a lean core + work-type profiles assembled by one setup script.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Check it deeper on your machine
It has changed since its scan on 11 Nov 2026.
As a Claude Code plugin, install it and then ask Claude Code:
/plugin marketplace add keithmackay/mackayi/plugin install scanskillsafety@mackayiDo a deep scan of https://github.com/PromptPartner/agentsmith with scanskillsafety
Or from a terminal, clone it and run it:
git clone https://github.com/keithmackay/scanskillsafety-skill ~/.claude/skills/scanskillsafetynode ~/.claude/skills/scanskillsafety/cli.cjs --deep 'https://github.com/PromptPartner/agentsmith'
It checks the exact commit: every file, its packages against OSV (api.osv.dev), how it compares with this page's scan (findsafeskills.com receives the repository's address), and, if you agree, Claude reading every file with your own Claude subscription. It never runs the code.
Badge for your README
Markdown
[](/listings/promptpartner--agentsmith)HTML
<a href="/listings/promptpartner--agentsmith"><img src="/api/listings/promptpartner--agentsmith/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
Choose one route. Both use the same guided setup and produce the same project files.
1. Manual setup
1. Download the standalone build for macOS, Windows, or Linux from the [latest release](https://github.com/PromptPartner/agentsmith/releases/latest). The macOS and Windows builds are currently unsigned, so your operating system may ask you to confirm before running them. 2. Open a terminal and run `agentsmith`. 3. Answer the guided questions. AgentSmith shows the exact plan before it writes anything.
The wizard can configure an existing project or create an empty folder and initialize Git for a new one. The standalone app includes its runtime; Python is not required.
2. Ask your agent
Paste this into Claude Code, Codex, or another coding agent:
Install AgentSmith for this project. Follow the official agent guide at https://github.com/PromptPartner/agentsmith/blob/master/AGENT-INSTALL.md Inspect the project first, explain your profile recommendation in plain language, show me the exact installation plan, and ask once before applying it.
The agent inspects the repository without running project code, recommends a profile from evidence, previews the managed files, installs AgentSmith, and runs the doctor check.
> Building AgentSmith from source is for contributors and advanced automation. See the [installation reference](INSTALL.md).
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 122 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.
- <div align="center">
- <img src="site/assets/promptpartner-mark-amber.svg" width="72" height="72" alt="PromptPartner mark">
- <p>
- <a href="https://github.com/PromptPartner/agentsmith/actions/workflows/verify.yml"><img alt="Verification" src="https://github.com/PromptPartner/agentsmith/actions/workflows/verify.yml/badge.svg?branch=master"></a>
- <a href="https://github.com/PromptPartner/agentsmith/releases/latest"><img alt="Latest release" src="https://img.shields.io/github/v/release/PromptPartner/agentsmith"></a>
- <a href="LICENSE"><img alt="MIT license" src="https://img.shields.io/badge/license-MIT-14213D.svg"></a>
- </p>
- <p>
- <strong>English</strong> · <a href="README.de.md">Deutsch</a> · <a href="README.es.md">Español</a> · <a href="README.fr.md">Français</a> · <a href="README.zh-CN.md">简体中文</a>
- </p>
- </div>
- # AgentSmith — the open-source harness for coding agents
- **Proof before done.**
- Move from vibe coding to agentic engineering. Give Claude Code or Codex your project rules,
- a definition of done, and checks that produce evidence. Understand the task, make a bounded
- change, verify the result, and carry decisions into the next session.
- Project rules and configured guardrails help address scope drift, repeated repair loops,
- and “done” claims without evidence.
- - **Start with the right problem.** Read existing code and define completion checks before editing.
- - **Keep changes reviewable.** Make small changes, choose native permissions, and enable optional secret-scanning hooks.
- - **Prove the result.** Run all configured checks, exercise the actual feature, and retain the evidence.
- - **Carry progress forward.** Save decisions and a clear handoff in your project.
- - **Improve from experience.** Capture a failure, propose a rule, hook, or test, and validate the change.
- 
- **Try the difference on one task.** [Run the disposable demo](docs/02-your-first-hour.md):
- reproduce a failure, make a bounded fix, and inspect the proof. Or
- [inspect the existing evidence bundle](docs/demos/first-verified-loop/README.md), including
- the failing test, verification, real-path output, receipt, and handoff/resume.
- Once your checks are trustworthy, explore [finite maker/checker runs](docs/21-autonomous-runs.md)
- with approved scope and explicit limits.
- ## Set up AgentSmith
- Choose one route. Both use the same guided setup and produce the same project files.
- ### 1. Manual setup
- 1. Download the standalone build for macOS, Windows, or Linux from the [latest release](https://github.com/PromptPartner/agentsmith/releases/latest). The macOS and Windows builds are currently unsigned, so your operating system may ask you to confirm before running them.
- 2. Open a terminal and run `agentsmith`.
- 3. Answer the guided questions. AgentSmith shows the exact plan before it writes anything.
- The wizard can configure an existing project or create an empty folder and initialize Git for a new one. The standalone app includes its runtime; Python is not required.
- ### 2. Ask your agent
- Paste this into Claude Code, Codex, or another coding agent:
- ```text
- Install AgentSmith for this project. Follow the official agent guide at
- https://github.com/PromptPartner/agentsmith/blob/master/AGENT-INSTALL.md
- Inspect the project first, explain your profile recommendation in plain language,
- show me the exact installation plan, and ask once before applying it.
- ```
- The agent inspects the repository without running project code, recommends a profile from evidence, previews the managed files, installs AgentSmith, and runs the doctor check.
- > Building AgentSmith from source is for contributors and advanced automation. See the [installation reference](INSTALL.md).
- ## New project or existing project?
- | Starting point | What AgentSmith does | What it does not do |
- |---|---|---|
- | Existing project | Inspects the files, recommends a profile, preserves foreign content, and adds managed rules and verification scaffolding. | It does not execute the project during inspection or replace existing project configuration. |
- | New project | Creates or validates an empty folder, initializes Git, and adds AgentSmith. | It does not choose or generate an application framework. Your agent can build the application after setup. |
- Project setup is the default because the rules travel with the repository and can be reviewed by collaborators. A user-wide core for every project is available under **Advanced options**. A layered setup combines that user-wide core with a small project-specific profile. See [project, user-wide, and layered setups](INSTALL.md#where-the-rules-live).
- ## Work profiles
- A profile tells the agent what “done” means for the work at hand. The wizard recommends one profile and shows the file evidence behind its choice. You can accept it or view the full list.
- | Profile | Use it for | Main proof |
- |---|---|---|
- | `software-dev` | Features, fixes, refactors, apps, libraries, and product UI | Build, checks, tests, security pass, real invocation |
- | `devops-setup` | Installers, CI, containers, configuration, and deployments | Dry run, idempotence, rollback, real deployment path |
- | `marketing-outreach` | Campaigns, email, newsletters, landing copy, and CRM work | Audience and factual review, links, rendering, send approval |
- | `document-creation` | Reports, proposals, specifications, manuals, and wikis | Source accuracy, structure, links, rendered file review |
- | `data-crunching` | Data cleaning, analysis, SQL, metrics, and ETL | Reproducibility, totals, edge cases, output inspection |
- | `general-admin` | Triage, scheduling, organization, summaries, and routine operations | Completeness, faithful output, destination and approval checks |
- | `deep-research` | Due diligence, market research, and cited investigations | Source quality, claim coverage, citations, synthesis review |
- | `creative-design` | Diagrams, decks, brand assets, images, and video | Brief, visual and export review, accessibility, brand consistency |
- | `security-audit` | Threat models, security reviews, penetration tests, and IAM audits | Reproduction, severity evidence, remediation, retest |
- `autonomous-loops` is an advanced modifier for scheduled or unattended work. Combine it with the main work profile; do not use it by itself. The [profile guide](docs/07-how-to-pick-a-profile.md) explains close calls, switching, and stacking.
- ## What gets installed
- - `AGENTS.md` is the canonical project agreement. Claude Code also receives a generated `CLAUDE.md` that imports it, so there is one rulebook.
- - `.harness/verify.conf` defines the real checks for this project.
- - `.agentsmith/state.json` records only AgentSmith-owned settings so updates and removal preserve foreign content.
- - Optional skills, MCP servers, and hooks are available in the collapsed advanced step.
- The cautious permission mode is the default. AgentSmith never treats a connected external service as permission to write to it.
- ## See the proof
- The [First Verified Loop](docs/demos/first-verified-loop/README.md) records a failing test before the fix, passing verification afterward, a real command invocation, a receipt, and a resumable handoff. The [support registry](config/agents.json) separates instruction support from tested native behavior.
- ## Documentation and community
- - [Installation reference](INSTALL.md)
- - [Documentation map](docs/README.md)
- - [How verification becomes evidence](docs/03-verify-means-evidence.md)
- - [Agent compatibility](docs/22-compatibility-contract.md)
- - [Contributing](CONTRIBUTING.md) and [support](SUPPORT.md)
- - [Security policy](SECURITY.md) and [code of conduct](CODE_OF_CONDUCT.md)
- MIT licensed. Built by [PromptPartner](https://promptpartner.ai/).
- Optional integrations disclose their effects before installation. See [capability selection](docs/local-capabilities.md).
- Recall local notes with `agentsmith memory search "topic"` or `agentsmith memory read docs/research/note.md`. Optional `--with-memory-startup` adds reference-only startup suggestions; see [project memory](docs/project-memory.md). Orca delivery requires host observation.
- `agentsmith audit-config` provides a read-only advisory report for selected project configuration. `--include-user` adds native user settings; `--fail-on high` adds a CI gate. See [configuration audit](docs/27-config-audit.md) for scope and limits.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.