OpenGAP
A Git-native CLI standard for defining, validating, and exporting portable AI agents.
- Source repo
- open-gitagent/opengap
- Stars
- ★ 3k
- Last updated
- 4mo ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 54/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformCodex · Claude Code · OpenAI API
- Cost
- Free software; you pay for model usage
- Setup effort
- Low · running in minutes
- You'll need
- Typical use
- An engineering team that wants to keep a code-review agent in Git and run opengap validate in CI on every change.
- Not a fit if
- Teams needing out-of-the-box orchestration, tool execution, or memory I/O
- Users who want a GUI instead of CLI and Git workflows
- Teams unwilling to depend on Node.js 18+, npm, and local filesystem access
- Source review
- 54/100 · Major gaps
What does this agent do, and when should you use it?
OpenGAP, or Git Agent Protocol, is a Git-native standard for agent definitions plus its reference CLI. A minimal agent consists of an agent.yaml manifest and a SOUL.md identity file, with optional rules, skills, tool schemas, workflows, memory, hooks, compliance artifacts, and recursive sub-agents. Its CLI scaffolds templates, validates definitions, displays metadata, imports selected formats, and exports targets including system prompts, Claude Code files, and OpenAI Agents SDK Python code. It captures the identity layer—prompts, rules, roles, model preferences, and tool schemas—while orchestration graphs, live tool execution, memory I/O, and iterative loops remain the responsibility of the destination runtime.
After installation, the opengap CLI (with gitagent retained as a compatibility alias) can scaffold minimal, standard, full, and llm-wiki definitions through opengap init --template. opengap validate checks a definition against the specification, while opengap validate --compliance checks regulatory and segregation-of-duties configuration; opengap audit produces a compliance checklist report. opengap export --format <fmt> turns repository definitions such as agent.yaml and SOUL.md into adapter-specific output, while opengap import --from <fmt> <path> supports claude, cursor, crewai, and opencode imports, and opengap run runs a definition from a Git repository or local directory through an adapter. YAML files in workflows/ can describe deterministic skill:, agent:, and tool: steps with depends_on ordering, templated inputs, conditions, prompts, and failure-notification settings.
- An engineering team that wants to keep a code-review agent in Git and run opengap validate in CI on every change.
- An agent owner who needs one reviewed set of identity files and tool schemas exported to Claude Code, OpenAI Agents SDK, or a generic system prompt.
- A financial-services team that needs to declare maker/checker conflicts, human supervision, recordkeeping, and audit controls in agent configuration.
- A platform team that wants to reuse a base agent, skills, or sub-agent definitions through extends and dependencies.
- A developer migrating the identity layer of a Claude, Cursor, CrewAI, or OpenCode agent into a versioned repository.
How do you install or deploy this agent?
Prerequisites are Node.js >= 18, npm, and Git. Install globally:
npm i -g @open-gitagent/opengapUse opengap after installation; gitagent is installed as a backward-compatible alias. No API credential is documented as a prerequisite for creating, validating, or exporting an agent definition.
How do you use this agent?
Create and inspect a first definition inside a Git repository:
opengap init --template standard
opengap validate
opengap info
opengap export --format system-promptKeep agent.yaml and SOUL.md for the minimum valid definition. To run a definition directly:
opengap run ./my-agent --adapter lyzrTo keep the agent definition separate from the target workspace:
opengap run --dir ./agents/reviewer --workspace ~/code/my-app --adapter claude -p "Review this repository"What are this agent's strengths and limitations?
- Uses commits, branches, diffs, tags, and pull requests as the lifecycle for agent definitions, enabling rollback and line-level auditability.
- Separates manifests, identity, hard rules, duties, skills, tools, and workflows into reviewable repository assets.
- Provides documented init, validate, audit, import, export, run, and skills commands alongside a broad adapter list.
- Includes explicit financial-compliance and segregation-of-duties structures, including role conflicts, required handoffs, and strict or advisory enforcement.
- Only agent.yaml is documented as having a strict schema; the quality and completeness of the other convention-based files remain a team responsibility.
- Exports do not carry over runtime orchestration, live tool execution, memory I/O, or iterative loops, so destination-runtime work is still required.
- The CLI requires Node.js >= 18, npm, Git, and local filesystem access; resolving Git dependencies and running from remote repositories also involves network access.
- The supplied documentation lists adapters but does not establish feature parity, authentication requirements, or production guarantees for each adapter.
How does this agent compare with similar options?
Relative to runtime frameworks named in the documentation—such as Claude Code, OpenAI Agents SDK, CrewAI, and LangGraph—OpenGAP focuses on extracting and versioning the identity layer, rules, roles, and tool schemas rather than replacing runtime orchestration or execution.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| OpenGAP This agent | 54 · Major gaps | CLIFree + model costs | ★ 3k | 4mo ago | TypeScript | Codex · Claude Code · OpenAI API |
| DACA Agentic AI Learning Stack | 37 · Major gaps | FrameworkFree + model costs | ★ 4.4k | 1y ago | Jupyter Notebook | OpenAI API |
| Vouch | 71 · Some gaps | CLIFree + model costs | ★ 66 | 3mo ago | Python | Claude Code |
| AnyAgent | 58 · Major gaps | Library / SDKFree + model costs | ★ 1.3k | 1mo ago | Python | — |
How does FollowAgents rate this agent?
Why each dimension lost points
The README describes .gitignore protection for .env, human-reviewed branches or PRs, role isolation, and rollback, giving some support for data-flow transparency and recovery. These are mostly standard and pattern descriptions; the supplied source does not show runtime permission enforcement, confirmation for writes, or secret handling, so least privilege, user confirmation, and sensitive-data handling score only 1. Workflows and dependency installation can have external effects, but consent and recovery details are not supplied. Dependencies are listed without security-review evidence. The publish workflow shows release permission configuration, but maintainer identity and update responsibility remain unclear.
The README, CLI entry point, package manifest, and workflows provide partial corroboration for commands and Node versions. Deductions reflect README options and aliases that cannot all be checked against the supplied source, and the CI example-validation step uses “|| true,” which can hide validation failures. Failure-message implementation and error-handling source are absent. Dependency names and version ranges are explicit and the minimum Node version is stated, but no lockfile or dependency-security evidence is included.
The README covers agent creation, import, export, compliance, workflows, and multiple runtime environments, while sample skills state intended scenarios and triggers. The porting section also bounds the capability by identifying the identity layer as portable and leaving runtime orchestration and tool execution in the original framework. Deductions reflect that scenarios and compatibility claims are mostly documentation; the supplied source is insufficient to verify specific adapter behavior or environment limits.
The README gives a clear directory structure, minimum required files, quick start, command table, and examples. The package and complete license text identify MIT. Deductions reflect the project's prior renames and inconsistent project or command naming in the README, no supplied changelog, and no clear maintainer or support commitment. Installation and examples are useful, but some commands and templates cannot be checked against the supplied source; limitations are mostly described as framework-porting boundaries.
The standard organizes portable agent definitions around Git files, history, and adapters; the README includes reusable configuration and report-format examples, supporting use in creation, review, and migration. Deductions reflect that compliance support and cross-framework export benefits are documented claims whose actual coverage cannot be established from the supplied files. Runtime, maintenance, human review, and compliance configuration also have costs that the material does not quantify or bound.
Many claims can be traced to README structures, commands, patterns, and examples, while the package manifest and CLI entry point corroborate the package name, version, and registered commands. Deductions reflect missing key materials such as the referenced specification, adapter implementations, and validator source; CI also permits example-validation failures to continue. The README presents compliance, adapter coverage, and workflow behavior in definite terms, but the supplied material does not consistently distinguish implemented facts from design goals.
- Treat compliance, permission isolation, and cross-framework support as documented design claims; the supplied material is insufficient to verify runtime enforcement.
- The CI example-validation command ignores failure status, so it does not establish that all examples pass.
Maintain this project? Add the FARS badge to your README
[](https://followagents.com/en/agents/opengap)<a href="https://followagents.com/en/agents/opengap"><img src="https://followagents.com/badge/opengap.svg" alt="FARS review"></a>Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
v0.5.0 v0.5.0 — Claude session continuation (--continue, --resume, --session-id)
Release notes excerpt
opengap run: session continuation for the claude adapter ```bash Continue the most recent session in the working directory opengap run -d ./my-agent -a claude -c -p "continue where we left off" Pin a session ID on first run, resume it later opengap run -d ./my-agent --session-id 3f9f74a6-6b7c-4a3e-9e1d-2f1f0a9b8c7d -p "Start the audit" opengap run -d ./my-agent --resume 3f9f74a6-6b7c-4a3e-9e1d-2f1f0a9b8c7d -p "Continue the audit" ``` New flags (claude adapter) - -c, --continue — continue the most recent Claude Code session in the working directory (pairs with --workspace) …View on GitHub ↗ -
v0.4.0 v0.4.0 — OpenGAP (rebrand from gapman)
Release notes excerpt
OpenGAP v0.4.0 — package and CLI renamed from gapman to opengap. Install: npm install -g opengap (or scoped @open-gitagent/opengap). The gitagent command stays as a backward-compat alias. Includes the financial_governance type fix and repository.url correction for provenance.View on GitHub ↗ -
v0.3.2 v0.3.2 — publish gapman both scoped and unscoped
Release notes excerpt
Publishes to both: - @open-gitagent/gapman (scoped) - gapman (unscoped alias, same contents) Either works: ```bash npm install -g gapman or npm install -g @open-gitagent/gapman ```View on GitHub ↗
FAQ
What is the minimum agent definition?
Can it target Claude Code or Codex?
Does it execute every workflow by itself?
Is a particular model or cloud provider required?
What are the license and documented costs?
Thanks — we use this to improve the page.