Skip to content

CLI Reference

SkillForge provides a CLI powered by Typer. Run skillforge --help to see all commands.

Commands

skillforge add <spec>

Fetch, security-audit, and install a skill directly from a GitHub repository into your active project, recording its SHA-256 digest in skillforge.lock:

skillforge add anthropics/skills:pdf
skillforge add gintingism/skillforge:systematic-debugging

Options: - -p, --project PATH: Target project root (defaults to .). - -g, --global: Install into user-level assistant configs. - --skip-audit: Skip security auditing before installation.

skillforge sync

Deterministically install and verify all skills pinned in skillforge.lock (verifying SHA-256 integrity and security audit rules before installing):

skillforge sync
skillforge sync --project ./my-repo

skillforge doctor

Run a full health, security, and context budget diagnostic across installed skills:

skillforge doctor
skillforge doctor --global
skillforge doctor --json

skillforge diff <target_a> [target_b]

Compare changes between a local skill and an upstream repo, or between two skill directories:

skillforge diff ./skills/my-skill
skillforge diff my-skill gintingism/skillforge:my-skill

skillforge budget

Profile context window token usage across installed or built-in skills, separating always-loaded startup pointer tokens from on-demand SKILL.md body and disclosed resource tokens:

skillforge budget
skillforge budget --builtin
skillforge budget --json

skillforge audit <path>

Scan a skill file or directory for prompt injections, zero-width unicode, dangerous shell commands, broken references/ or scripts/ pointers, and progressive disclosure token limits:

skillforge audit ./skills/my-skill
skillforge audit ./skills/my-skill --strict
skillforge audit ./skills/my-skill --sandbox
skillforge audit ./skills/my-skill --sarif findings.sarif
skillforge audit ./skills/my-skill --json

skillforge audit-mcp [path]

Audit an MCP configuration file (mcp.json or claude_desktop_config.json) for hardcoded secrets, shell wrappers, unpinned packages, and overbroad directory grants:

skillforge audit-mcp ./mcp.json
skillforge audit-mcp ./mcp.json --strict
skillforge audit-mcp ./mcp.json --json

skillforge list

List all installed skills across detected assistants:

skillforge list
skillforge list --global

skillforge remove <name>

Remove an installed skill from all detected assistants and update skillforge.lock:

skillforge remove my-skill

skillforge create <name>

Create a new skill file or modular directory structure:

skillforge create code-review -d "Review code carefully. Use when auditing PRs."
skillforge create full-tool -d "Complete tool. Use when automating workflows." --full

skillforge validate <path>

Validate a SKILL.md file against schema standards:

skillforge validate ./skills/my-skill/SKILL.md
skillforge validate ./skills/my-skill/SKILL.md --json

skillforge pack <path> / skillforge unpack <archive>

Package a skill directory into a zip file or safely unpack one:

skillforge pack ./skills/my-skill --output ./my-skill.zip
skillforge unpack ./my-skill.zip --output ./skills

skillforge search <query>

Search for skills across local and built-in registries:

skillforge search debugging

skillforge install <name>

Install a local or built-in skill into detected assistant targets:

skillforge install systematic-debugging
skillforge install systematic-debugging --global

# With custom security audit gate score
skillforge install custom-skill --min-score 85

# Bypass security gate (insecure)
skillforge install legacy-skill --insecure

skillforge search-remote [query]

Search the remote community skills catalog:

skillforge search-remote "trifecta"
skillforge search-remote "" --tag security

skillforge index [repo_dir]

Scan and index all skills from a repository into the local discovery cache:

skillforge index ./community-skills --url https://github.com/org/repo

skillforge lsp

Start the Language Server Protocol (LSP) daemon over standard I/O:

skillforge lsp

skillforge keygen

Generate a high-entropy 256-bit signing key for skill releases:

skillforge keygen --out skill.key

skillforge sign <path>

Sign a skill directory or skillforge.lock with an HMAC-SHA256 manifest signature:

skillforge sign ./skills/my-skill --key-file skill.key --signer "release-bot"
skillforge sign ./skillforge.lock --key-file skill.key

skillforge verify-signature <path>

Verify that all files in a skill or lockfile match their signed digests:

skillforge verify-signature ./skills/my-skill --key-file skill.key
skillforge verify-signature ./skillforge.lock --key-file skill.key

skillforge tui

Launch the terminal user interface dashboard:

skillforge tui --static
skillforge tui --watch

skillforge firewall-mcp

Run the inline zero-trust MCP JSON-RPC security firewall or inspect a JSONL traffic capture:

skillforge firewall-mcp --audit-log ./mcp-audit.jsonl -- npx -y @modelcontextprotocol/server-postgres
skillforge firewall-mcp --check ./captured-traffic.jsonl --json

skillforge bench [path]

Run the multi-model semantic trigger benchmark matrix (Precision, Recall, F1, False Positive Rate):

skillforge bench ./skills/my-skill
skillforge bench . --output ./benchmark_scorecard.json

skillforge refactor-progressive <path>

Slice a monolithic SKILL.md into progressive disclosure references with semantic preservation verification:

skillforge refactor-progressive ./skills/bloated-skill --target-tokens 500

skillforge run-sandboxed <script>

Execute a bundled Python script inside an isolated zero-trust capability jail:

skillforge run-sandboxed ./skills/my-skill/scripts/helper.py --timeout 5.0