A3S-Lab

Prompt Injection Triage — Development skill for Claude Code

Development community

Assess whether flagged I/O text is a real prompt-injection / jailbreak that warrants action.

How to install Prompt Injection Triage

Installs to ~/.claude/skills/a3s-lab-sentry-prompt-injection-triage/SKILL.md

Terminal

mkdir -p ~/.claude/skills/a3s-lab-sentry-prompt-injection-triage && curl -fsSL https://raw.githubusercontent.com/A3S-Lab/Sentry/HEAD/skills/prompt-injection-triage.md -o ~/.claude/skills/a3s-lab-sentry-prompt-injection-triage/SKILL.md

Restart Claude Code, or start a new session, for it to be picked up.

What Prompt Injection Triage does

name: prompt-injection-triage description: Assess whether flagged I/O text is a real prompt-injection / jailbreak that warrants action.

Prompt-injection triage

Use this when L1/L2 flagged `SslContent` (the agent's LLM prompt or completion) as containing injection or jailbreak markers.

What to establish

Direction. Is the marker in the agent's input (a tool result, a fetched web page, a file it read — i.e. untrusted content trying to hijack the agent) or its output? I

Alternatives in Development

Dashclaw Quality — Recurring find-and-fix quality pass over the whole DashClaw app — browser smoke (frontend-verify) + code gates 297 ★
01 Fts5 Search — Replace Inkwell's naive LIKE post filter with a real SQLite FTS5 full-text index that ranks 143 ★
Fable Jailbreak — Anthropic's Fable jailbreak for Claude Code CLI 136 ★