All authors
Claude Skills by howardhsieh
github.com/howardhsieh14 skills0 installs10 views
- GuardShows the status of the agentsec-guard runtime guardrails, explains why a tool call was blocked or questioned, dry-runs a command against the rules, and approves the currently installed plugins and skills as the new baseline after a review. Use when the user asks about agentsec-guard, a G001-G005 or S001 message, or wants to approve installed packages.Votes: 0GitHub stars: 2
- Agent Config AuditAudits local AI coding-agent configuration (Claude Code settings, permissions, hooks, sandbox and MCP servers; OpenAI Codex config.toml; Cursor mcp.json and sandbox.json) for risky settings, leaked secrets and repository-supplied code, then proposes a prioritized hardening plan with ready-to-paste snippets. Use when setting up a machine or repo, before opening an untrusted repository in an agent, for periodic hygiene, or after an incident.Votes: 0GitHub stars: 2
- Agent Incident ResponseGuides incident response when an AI agent setup may be compromised, such as a malicious or hijacked skill, plugin or MCP server, a prompt-injected agent that took harmful actions, or a leaked token the agent could reach. Covers triage, evidence preservation, human-approved containment, credential rotation, scoping via transcripts and persistence checks, eradication across agents, and the report. Use for agent incident, compromised skill, malicious plugin, leaked token.Votes: 0GitHub stars: 2
- Agent Security CheckupRuns a one-command security checkup of the user's whole AI coding-agent setup - Claude Code, Codex and Cursor settings, every installed skill and plugin across agents, and MCP servers - and grades it A to F with a shareable HTML report, top risks and quick-win fixes. Use when the user asks for a security checkup, health check, security score or grade, to audit their agent setup, or runs /checkup.Votes: 0GitHub stars: 2
- Agent Threat ModelThreat-models LLM and AI agent systems such as coding-agent setups, support agents, RAG apps with tools, multi-agent systems and MCP integrations. Inventories tools, credentials and trust boundaries, draws a data-flow diagram, checks each agent context for the lethal trifecta, and maps threats to OWASP Agentic Top 10, LLM Top 10 and MITRE ATLAS with rated mitigations. Use for a threat model, STRIDE for agents, prompt injection risk, lethal trifecta, or design review of an agent.Votes: 0GitHub stars: 2
- Agent Trace DetectionBuilds and runs detections over AI coding-agent activity - normalizes Claude Code session transcripts and OpenTelemetry exports, runs bundled Sigma rules (for SIEMs) and TraceSig rules (provenance-aware, e.g. secret read then network call, untrusted content then publish), and guides writing and testing new agent detections. Use to set up monitoring for coding-agent fleets, hunt through local transcripts after a suspicious session, or write detection rules for agent tool calls.Votes: 0GitHub stars: 2
- Mcp Server Security ReviewReviews MCP server source code for security flaws - tool poisoning and rug-pull risk in model-facing text, command injection, path traversal, SSRF, token passthrough, confused deputy, weak MCP auth and secret leaks - and reports verified findings with fixes mapped to the MCP 2026-07-28 spec. Use for an MCP server security review, to audit MCP server code, to check MCP auth, or before adopting an MCP server from a third party.Votes: 0GitHub stars: 2
- Skill Supply Chain AuditReviews an Agent Skill, Claude Code plugin, marketplace repo or bundled MCP config before it is installed - inventories hooks, scripts, MCP servers and hidden text, flags download-and-execute, credential access, persistence and covert instructions, then pins the reviewed version with a hash lockfile and re-reviews only what changed on update. Use before installing any third-party skill or plugin, when a pinned one updates, or for a periodic skill inventory.Votes: 0GitHub stars: 2
- Pdf Helper ProFIXTURE ONLY. Inert stand-in for a suspicious third-party skill used by the agent-incident-response tests. Converts PDF files.Votes: 0GitHub stars: 2
- Pdf Helper ProFIXTURE ONLY. Inert stand-in for a suspicious third-party skill used by the agent-incident-response tests. Converts PDF files.Votes: 0GitHub stars: 2
- Benign NotesFIXTURE ONLY. Benign skill that must never match the suspect.Votes: 0GitHub stars: 2
- PdfhelperFIXTURE ONLY. Differently named skill that reuses the suspect helper script.Votes: 0GitHub stars: 2
- Benign SkillFormats CSV files into Markdown tables. Use when the user asks to turn a CSV into a table.Votes: 0GitHub stars: 2
- Pdf HelperExtracts text from PDF files. Use when the user shares a PDF.Votes: 0GitHub stars: 2