Score Agent — Development skill for Claude Code
Score captured agent traces with the 4 detection arms (A taint, B unrequested action, C output echo, D provenance) and apply both gates.
How to install Score Agent
Installs to ~/.claude/skills/elementalsouls-claude-agenthunter-score-agent/SKILL.md
mkdir -p ~/.claude/skills/elementalsouls-claude-agenthunter-score-agent && curl -fsSL https://raw.githubusercontent.com/elementalsouls/Claude-AgentHunter/HEAD/commands/score-agent.md -o ~/.claude/skills/elementalsouls-claude-agenthunter-score-agent/SKILL.md Restart Claude Code, or start a new session, for it to be picked up.
What Score Agent does
name: score-agent
description: Score captured agent traces with the 4 detection arms (A taint, B unrequested action, C output echo, D provenance) and apply both gates. A target is clean only when all arms agree AND their controls fired. Usage: /score-agent [--baselines d1,d2]
Run the deterministic engine over traces the user points to:
- Normalise traces via
engine/adapters/into the single Trace shape. - `python3 engine/score.py agentdojo
--bas
Alternatives in Development
- Check 1: Cache Exists? — CLAUDE_DIR="${CLAUDE_CONFIG_DIR:-$HOME/.claude}" CACHE_EXISTS=$(ls -d "$CLAUDE_DIR/plugins/cache/claude-hud" 2 10.9k ★
- Echo Analyst — Use the @echo-analyst agent to surface requirement gaps and assumptions for the following: $ARGUMENTS 523 ★
- Lacp — Local-first policy, evidence, memory, and recovery controls for Claude, Codex, Hermes, and other CLI coding ag 300 ★
Full documentation available on GitHub
View Source RepositoryRelated Skills
Technical Cofounder
Your technical cofounder and his senior dev team: a fleet of variable agent loadouts with complementary suppor
Pensieve
Your coding agent has amnesia and tunnel vision. Pensieve is the context engine that fixes both — long-term me
Fuzz Validate
Prove the fuzz harness actually reaches the target (poison-pointer + under-alloc controls)
Livewire Alpine Skills
Agent Skills for Laravel Livewire v4 + Alpine.js v3. Recipes EXECUTED against a real install (14 tests), a 21-
Kimi Guard
Runtime behavior guard for Kimi Code CLI — loop detection, quota gates, checkpoint/resume, supervised Wire run
Axguard Flow
Dataflow / taint path analysis (diagnostic). Usage: /axguard-flow [path]
Related Agents
Lead Judge
Use to work the Lead Radar queue: read each lead, agree or disagree with its score, approve the real ones with
Product QA Reviewer
Critical quality gatekeeper at both ends of the delivery: before implementation, runs the readiness check (Def
Android Mariana Hunter
Runs Mariana Trench interprocedural taint analysis on an Android APK and turns coherent source-to-sink traces