Repo Agent Instruction Security Scan v1.2.0

What changed

• Scans every SKILL.md in addition to AGENTS.md, CLAUDE.md, Gemini, Cursor, Copilot, Claude, and Windsurf instruction files.
• Applies the existing eight deterministic review signals to Agent Skills without executing their instructions or bundled code.
• Preserves the 150 KB per-file limit, 200-file cap, dependency-directory exclusions, symlink exclusion, and workspace path boundary.
• Updates the GitHub Action, pre-commit hook, CLI documentation, and regression suite for the expanded instruction surface.

Verification

• All seven deterministic tests pass.
• The public release workflow smoke-tested the packaged CLI and pre-commit integration.
• A fresh public-asset install detected a high-severity download-and-execute chain inside skills/review/SKILL.md.
• The release archive carries signed GitHub build provenance.

Verify the downloaded archive before execution:

gh attestation verify repo-agent-instruction-security-scan-1.2.0.tar.gz -R sunxiayi/repo-agent-instruction-security-scan

SHA-256: d47c96668525357f60d0b9528118bc668f8eaabf7e2fd0fe87809536115851da

Repo Agent Instruction Security Scan v1.1.3

This release is built from the tagged source by GitHub Actions and carries signed build provenance.

Verify the downloaded archive before execution:

gh attestation verify repo-agent-instruction-security-scan-1.1.3.tar.gz -R sunxiayi/repo-agent-instruction-security-scan

The adjacent .sha256 file provides the exact archive digest used by Homebrew and Scoop.

Repo Agent Instruction Security Scan v1.1.2

Adds independently validated pre-commit integration and makes the immutable v1.1.2 tag the documented rev. CI validates the hook manifest and runs the Node hook through pre-commit 4.6.2 before every release.

Install the same verified source archive through a standard package manager:

# macOS or Linux
brew install sunxiayi/tap/repo-agent-scan

# Windows
scoop bucket add repoagentkit https://github.com/sunxiayi/scoop-bucket
scoop install repoagentkit/repo-agent-scan

The Homebrew formula and Scoop manifest both pin the v1.1.2 archive with SHA-256 063baecb97197d05c8324883f44f25f1f8c82c1ca5f3aba7301a68e4e44e8b10. The scanner remains zero-dependency, read-only during scanning, network-free during scanning, and telemetry-free. Package SHA-256: e99ad514e8c03aa82eb5bdf6213eb8cf1d75c1426e0ccd377e1643e7d236ea9c.

Repo Agent Instruction Security Scan v1.1.1

Recommended CLI release. Adds a zero-dependency local scanner, SARIF 2.1.0 reports, and a pre-commit hook while preserving the read-only GitHub Action.

Run the reviewed release directly from GitHub:

npx --yes github:sunxiayi/repo-agent-instruction-security-scan#v1.1.1 .

Or install the versioned release package:

npm install --global https://github.com/sunxiayi/repo-agent-instruction-security-scan/releases/download/v1.1.1/repo-agent-instruction-security-scan-1.1.1.tgz

Package SHA-256: 5485ede739bc4f5c06cd117678741e4987688765517032f304139aacd99cd35b

No account or token is required. The scanner executes no repository instruction content, makes no network requests during scanning, and collects no telemetry. Static findings are review prompts rather than vulnerability verdicts.

Repo Agent Instruction Security Scan v1.1.0

Adds a zero-dependency local CLI, SARIF 2.1.0 reports, and a pre-commit hook while preserving the existing read-only GitHub Action.

Run directly from GitHub:

npx --yes github:sunxiayi/repo-agent-instruction-security-scan -- .

The CLI and Action execute no repository instruction content, make no network requests during scanning, require no token, and collect no telemetry. Static findings remain review prompts rather than vulnerability verdicts.

Repo Agent Instruction Security Scan v1.0.0

First stable release.

• Scans AGENTS.md, CLAUDE.md, GEMINI.md, Cursor, Copilot, Claude, and Windsurf instruction files.
• Emits line-level annotations, a job summary, JSON output, and severity counts.
• Uses eight deterministic early-warning checks with no dependencies, network calls, code execution, token, or telemetry.
• Runs on Node 24 and fails on high-severity findings by default.

Findings are review prompts, not proof of malicious intent; a clean result is not a security guarantee.