Agent Skill

Dalfox ships a SKILL.md that teaches any skill-aware agent how to drive Dalfox correctly — authorization gate first, MCP over CLI, preflight before heavy scans, V > A > R finding prioritization. Point your agent at it and the model stops guessing flags.

The file lives at skills/dalfox/SKILL.md in the repository, so once you've cloned Dalfox you already have it locally.

Install with npx skills

The easiest way is the open agent-skills CLI:

# Install to the current project (committed, shared with the team)
npx skills add hahwul/dalfox

# Or install globally (~/<agent>/skills/, available everywhere)
npx skills add hahwul/dalfox -g

The CLI auto-detects which agents you use (Claude Code, Cursor, Codex, OpenCode, and ~45 others) and links the skill for each. Pick a specific agent if you only want one:

# Only Claude Code
npx skills add hahwul/dalfox -a claude-code

# Non-interactive (CI-friendly)
npx skills add hahwul/dalfox -g -a claude-code -y

To update later:

npx skills update dalfox

To remove:

npx skills remove dalfox

Install manually

If you prefer not to run npx, clone the skill file into the location your agent expects. For Claude Code that's ~/.claude/skills/dalfox/SKILL.md:

mkdir -p ~/.claude/skills/dalfox
curl -o ~/.claude/skills/dalfox/SKILL.md \
  https://raw.githubusercontent.com/hahwul/dalfox/main/skills/dalfox/SKILL.md

Other clients read from their own skills directory — see your agent's docs for the exact path.

What the skill covers

  • Trigger conditions — fires when the user asks to scan a URL for XSS, enumerate reflected parameters, or mentions "dalfox" explicitly. Skips non-XSS vulnerabilities.
  • Authorization gate — the skill refuses to scan until the user confirms they're authorized to send payloads to the target.
  • Mode detection — prefers MCP tools when available, falls back to the dalfox CLI, and tells the user how to install if neither is present.
  • MCP playbook — preflight_dalfox → scan_with_dalfox → poll get_results_dalfox (honoring suggested_poll_interval_ms) → delete_scan_dalfox when done. Includes the validated input bounds (timeout 1–299 s, delay 0–9999 ms) so the agent doesn't send values Dalfox will reject.
  • CLI scenarios — POST bodies, authenticated sessions, proxy-through-Burp, blind XSS with a callback URL, stored XSS, pipe input, fast smoke tests, maximum-coverage runs, and machine-readable output.
  • Result interpretation — finding type legend (V verified DOM execution > A AST-detected > R reflected-only) so the agent leads with confirmed hits.
  • Failure modes — what reachable: false, all-R findings, a stuck scan, or an invalid_params response actually mean, and how to recover.

Prerequisite

The skill assumes the dalfox binary or MCP server is reachable from the agent's environment. Install Dalfox first (installation guide), then install the skill.

Authoring tips

The skill file lives at skills/dalfox/SKILL.md in the repository. Contributions are welcome — keep it focused on how an agent should drive Dalfox, not on restating CLI reference material that already lives in the CLI reference.

ESC