by tang-vu
Maintainer-governed agent for evidence-backed open-source contribution proposals
# Add to your Claude Code skills
git clone https://github.com/tang-vu/ContribAILast scanned: 5/30/2026
{
"issues": [],
"status": "PASSED",
"scannedAt": "2026-05-30T15:25:12.869Z",
"npmAuditRan": true,
"pipAuditRan": true
}See how ContribAI compares with popular alternatives.
ContribAI is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by tang-vu. Maintainer-governed agent for evidence-backed open-source contribution proposals. It has 244 GitHub stars.
Yes. ContribAI passed SkillsLLM's automated security scan — a dependency vulnerability audit plus prompt-injection heuristics — with no high-severity issues. You can read the full report in the Security Report section on this page.
Clone the repository with "git clone https://github.com/tang-vu/ContribAI" and add it to your Claude Code skills directory (see the Installation section above).
ContribAI is primarily written in Rust. It is open-source under tang-vu on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh ContribAI against similar tools.
No comments yet. Be the first to share your thoughts!
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM is an informational directory and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
A maintainer-governed admission and evidence layer for AI-assisted open source contributions.
Website · Deploy website · Why ContribAI · Safety model · Quick start · Consent protocol · Architecture
ContribAI analyzes repositories and prepares small, reviewable changes. It does not treat a public repository as permission to write. Submission is a separate capability that must be enabled by the operator, authorized by the target maintainer, bounded by policy, reviewed by a human, and published as a draft pull request with an evidence receipt.
The project website is a dependency-free introduction and read-only onboarding path. It is informational only and exposes no ContribAI runtime capability.
The Vercel deployment serves only the static files under site/. It needs no environment
variables, tokens, functions, or access to the credential-bearing Rust runtime. Project
maintainers should follow the existing-repository deployment guide
instead of cloning the repository through the button.
public code ≠ permission to submit
generation ≠ admission
passing checks ≠ maintainer approval
AI can make code generation cheap while making review expensive. The failure mode is not merely bad code; it is uninvited work that transfers verification cost to maintainers. ContribAI is built around the opposite incentive:
This is a tool for maintainers and accountable contributors—not a contribution-farming bot.
An external draft can be created only when every gate passes:
operator --submit
│
▼
repository manifest OR maintainer issue label
│
▼
24-hour ContributionPermit bound to repository + base SHA
│
▼
scope, protected-path, size, risk, quality, and validation checks
│
▼
interactive human review
│
▼
draft PR + EvidenceCapsule
Important invariants:
| Invariant | Enforcement |
|---|---|
| No implicit write access | Pipeline write capability defaults to off |
| No unsolicited proposal | Repository consent or approved issue label required |
| No moving-base ambiguity | Contribution branch starts at the attested commit SHA |
| No review/evidence substitution | Full candidate fingerprint and scope are recomputed at the write boundary |
| Revocable until write | Current manifest or issue approval is re-read immediately before the first write |
| No governance takeover | License, policy, workflow, funding, CODEOWNERS, and agent instruction paths are protected |
| No silent publication | GitHub PR creation always sets draft: true |
| No delegated legal act | CLA signing is never exposed through CLI automation or MCP |
| No hidden MCP mutation | MCP advertises read-only tools unless explicitly started with --allow-writes |
See the threat model and the consent protocol for precise boundaries.
curl -fsSL https://raw.githubusercontent.com/tang-vu/ContribAI/main/install.sh | bash
On Windows PowerShell:
irm https://raw.githubusercontent.com/tang-vu/ContribAI/main/install.ps1 | iex
The installers select the latest published stable release and verify its checksum.
Set CONTRIBAI_VERSION to an exact tag (for example, v6.10.0) to pin a release.
See installation options for platform commands.
The isolated install path is smoke-tested against the published binary on Linux, macOS, and Windows for every release.
Requirements: Rust stable and Git. Credentials are not needed for the offline proof.
git clone https://github.com/tang-vu/ContribAI.git
cd ContribAI
cargo install --path crates/contribai-rs --locked
Run the production consent, admission, and evidence path without config, credentials, network access, or external writes:
contribai demo
contribai demo --json
contribai demo --manifest examples/quickstart-repository/.github/contribai.yml --json
The demo also probes a protected workflow path and confirms that it is denied. Continue with the five-minute walkthrough and inspect the bundled quickstart repository.
Configure secrets through environment variables instead of committing them:
export GITHUB_TOKEN="..."
export GEMINI_API_KEY="..." # or OPENAI_API_KEY / ANTHROPIC_API_KEY
contribai init
contribai doctor
These commands do not submit anything:
contribai analyze https://github.com/owner/repo
contribai target https://github.com/owner/repo
contribai run
contribai patrol
contribai mcp-server
To run the v7 contribution lifecycle against a maintainer-authorized issue:
contribai contribute owner/repo --issue 123 --dry-run
contribai contribute owner/repo --issue 123
contribai runs
contribai inspect-run <run-id>
contribai conformance
--dry-run executes through evidence packaging and stops: no review prompt, no external write.
conformance runs the offline safety-invariant suite. runs and inspect-run are read-only.
To request an admitted draft proposal:
contribai target https://github.com/owner/repo --submit
contribai solve https://github.com/owner/repo --submit
contribai contribute owner/repo --issue 123 --submit
--submit is necessary but not sufficient. The repository must opt in, every check must pass, and
the local operator must approve the exact evidence-bearing change interactively. The review renders
every regular and test change without truncation; current consent is checked again before writing.
Maintainers can opt in repository-wide with .github/contribai.yml:
schema_version: 1
enabled: true
max_files: 3
max_changed_lines: 120
allowed_paths: src/**, tests/**
The uppercase marker paths remain supported for compatibility with the experimental v1 protocol.
Operators can inspect the gate without invoking an LLM or writing to GitHub:
contribai consent-check owner/repo
contribai consent-check owner/repo --json --require-consent
For narrower approval, apply one of these labels to an issue:
contribai-approved (canonical)agent-ready or ai-contribution-approved (compatibility aliases)Issue approval is scoped to that issue. Repository consent is still bounded by its path and size budgets. Missing, malformed, or disabled consent is a denial.
Each admitted draft includes a compact receipt such as:
Permit: 1d28…
Consent: maintainer label `agent-ready` on #42
Base revision: 7a93…
Change fingerprint: c441…
Scope: 2 files / 37 changed lines
Evidence expires: 2026-08-11T08:00:00Z
Submission mode: draft only
Checks: admission, quality, risk, validation
The capsule makes the proposal reproducible and reviewable. The write boundary rejects expired, failed, cross-repository, scope-mismatched, or fingerprint-mismatched capsules and revalidates live maintainer consent. It remains a local audit receipt, not a substitute for CI, code review, provenance attestation, or maintainer judgment.
ContribAI attempts to append every terminal admission decision to the local audit ledger. Records contain the candidate fingerprint, permit/base SHA when available, scope, checks, decision stage, and reason—never generated file contents. Each record's SHA-256 receipt binds the preceding receipt, so accidental edits or broken ordering are detectable when the complete local chain is verified:
contribai admissions
contribai admissions --repository owner/repo --decision blocked
contribai admissions --json
The hash chain is a local integrity check, not a signature or remote attestation. An approved submission fails closed if its audit decision cannot be persisted. Each append verifies the retained history in the same write transaction; corrupted history blocks new receipts. See audit recovery for preserving and restoring evidence.