create-sub-agent — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited create-sub-agent (Agent Skill) and scored it 87/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 3 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 3 flagged
The text {match} tells the agent to skip the normal "ask the user first" gate. Used adversarially it removes the human-in-the-loop check before destructive or sensitive actions, turning a normally-gated agent into a fire-and-forget executor.
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
Follow the create-sub-agent skill workflow to design and generate a Claude Code agent file.
$ARGUMENTS — optional agent name or brief use-case description passed as initial contextto the design interview. Omit to start with open discovery.
$ARGUMENTS is provided, use it as the starting context for agent name / purpose(purpose, input/output contract, escalation posture, tools, permissions.deny, model, maxTokens, color, lifecycle hooks, placement), present design summary, confirm, then generate the agent .md file
validate_agent.py.github/agents/ or .claude/agents/ and ensure the path is tracked in .gitignore.Agent .md file with complete YAML frontmatter (name, description with <example> blocks, model, maxTokens, color, permissions.allowedTools, permissions.deny) and a second-person system prompt targeting 500-3,000 characters.
Plugin agents: flat .md file — plugins/<plugin-name>/agents/<agent-name>.md
skills/<name>/SKILL.md subdirectory format, but agents do NOT.feature-dev, code-simplifier, hookify).Local/project agents: .claude/agents/<agent-name>.md (also flat, no subdirectory).
To make an agent visible to GitHub Copilot or Claude Code for this repository, follow the Discovery-First Publication pattern:
.github/agents/ (Copilot) or .claude/agents/ (Claude)..gitignore allows the specific agent file or subdirectory.See Agent Discovery and Publication Pattern for details.
$ARGUMENTS is empty: conduct the full Phase 1 design interview — do not pre-fill~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.