Skill Install — Secure Skill Installer
Installs Claude skills from any GitHub repository with automated pre-install security scanning that blocks malicious code and backdoors.
The skill designs a layered security flow: scan first, graded outcomes (SAFE/WARNING/DANGEROUS), mandatory user confirmation on WARNING, non-overridable rejection on CRITICAL, and sandbox reminders; data flow (GitHub API → raw content → write to ~/.claude/skills) is clearly disclosed. Deductions: the security scan is an LLM prompt template rather than deterministic detection; no integrity verification (hash/signature), no rollback or uninstall mechanism, writes into the user's home directory without least-privilege constraints, and no isolation guarantee for fetched third-party content.
Workflow steps are self-consistent; URL parsing, API endpoints and raw-content paths are concrete, with clear examples. Deductions: fully dependent on WebFetch and GitHub API availability; failures such as network errors, API rate limits, empty file trees, or invalid frontmatter get only a rough Step 8 summary, lacking explicit failure feedback, retry or diagnostic guidance; key paths unverifiable in static review.
Triggers are explicit (GitHub URL provided, install-skills request, browsing repository skills), audience clear, examples cover typical scenarios, multi-select interaction is sensible. Deductions: non-fit boundaries undeclared (private repos, non-standard layouts, very large repos); core function depends entirely on api.github.com and raw.githubusercontent.com, which may be unreachable from mainland-China networks, with no mirror or fallback provided.
Clear structure with good progressive disclosure (SKILL.md + references/security_scan_prompt.md), standard frontmatter, a scan template with output format and examples; repo-level AGPL-3.0 license, version 6.7.0, CI and release workflows. Deductions: no per-skill version/changelog, no named maintenance responsibility, no FAQ or known-limitations section.
Goal (securely installing skills from GitHub) is clear; outputs (install summary, skip reasons) are directly usable, with marginal value over manual copying. Deductions: scan effectiveness rests entirely on LLM judgment with no verifiable accuracy; false-positive/negative risk unquantified; installation alone is roughly equivalent to clone-and-copy, so the main added value (scanning) lacks supporting evidence.
Auditable primary material exists (full workflow, scan template, three example outputs); the repo has CI and codeagent-wrapper tests, but these do not cover skill-install's key paths. Deductions: no tests or third-party execution evidence for this skill; per calibration, verifiability cannot exceed 5 on a static read.
- The security scan is LLM-prompt-driven, not deterministic detection, and cannot serve as a reliable defense against malicious code; independently review third-party skills before installing.
- Core function depends on api.github.com and raw.githubusercontent.com, which may be unreachable from mainland-China networks; no mirror or fallback is provided.
- No rollback/uninstall mechanism; installed skills run with Claude's permissions and must be removed manually from ~/.claude/skills/.
- This is a static review; nothing was executed and all runtime conclusions are inferences.
What does this skill do, and when should you use it?
skill-install is one of the standalone skills in the stellarlinkco/myclaude repository. It browses the skills/ directory of any GitHub repo via the GitHub API, lets the user multi-select skills to install, downloads every file, and runs a seven-category security analysis covering malicious command execution, backdoors, credential theft, and more, before producing a SAFE/WARNING/DANGEROUS verdict. Approved skills are installed to ~/.claude/skills/, and skills with critical findings are refused outright with no override. Install it individually through the repository's interactive npx installer.
Parses a user-supplied GitHub repository URL (repo root or /skills subdirectory formats); calls the GitHub API to list available skill directories; presents a multi-select prompt of skills to install; downloads each selected skill's SKILL.md, scripts/, references/, and assets/ file by file; applies the security scan template from references/security_scan_prompt.md and outputs a security status, risk level, detailed findings, and an APPROVE/REJECT recommendation; writes approved skills to ~/.claude/skills/{skill_name}/ preserving directory structure and executable permissions; finally summarizes installed and skipped skills with reasons.
- A Claude Code user wanting to try community skill repositories who wants a security review before copying files blindly
- A team admin distributing a subset of skills from an internal repo, letting members pick which ones to install
- A security-conscious user who downloaded a skill of unknown origin and wants to check for credential theft such as reading ~/.ssh, ~/.aws, or environment variables
- A developer maintaining a personal skill library who wants a standard browse→select→scan→install flow into ~/.claude/skills/
- Anyone installing a skill with borderline behavior (e.g., HTTP requests) who wants an explicit warning and a confirm/deny choice
What are this skill's strengths and limitations?
- Mandatory seven-category pre-install security scan covering malicious command execution, backdoors, credential theft, privilege escalation, and supply-chain attacks
- CRITICAL findings block installation with no override option, preventing accidental installation of high-risk skills
- WARNING findings are surfaced to the user for an explicit decision, balancing safety and flexibility
- Supports batch multi-select installation with a summary of installed and skipped skills
- The analysis template lives in references/security_scan_prompt.md, so the judgment criteria are auditable
- The security scan is a model-driven heuristic based on a prompt template — not a substitute for real static analysis or sandboxed execution
- Explicitly depends on the WebFetch and AskUserQuestion tools, which require adaptation in other clients
- No dedicated test suite or independent version notes are provided for this skill
- Licensed under AGPL-3.0; commercial closed-source use requires contacting the support email for a commercial license
How do you install this skill?
Install the collection via the interactive installer: npx github:stellarlinkco/myclaude. Run npx github:stellarlinkco/myclaude --list to see individually installable items, including skill-install. After installation the skill lives at ~/.claude/skills/skill-install/. No standalone command for installing only this one skill is documented.
How do you use this skill?
Trigger it in Claude Code with a prompt such as "Install skills from https://github.com/example/claude-skills" or "install skills from GitHub". The skill parses the URL, lists available skills for multi-select, downloads and security-scans them, asks for confirmation where warnings are found, and installs approved skills to ~/.claude/skills/. You may need to restart Claude or use the skills directly afterwards.
How does this skill compare with similar options?
Compared with manually git-cloning a repo and copying folders into ~/.claude/skills/, this skill adds a model-driven pre-install security review and batch skill selection; the repository itself names no competing alternatives.