一键在 Manus 中运行任何 Skill

ctf

星标6

分支0

更新时间2026年3月31日 12:23

Start CTF challenge solving pipeline. Auto-matches "ctf", "solve challenge", "pwn", "reversing", "crypto challenge", "wargame"

安装

用 Codex 或 Claude 帮你安装复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。

在 Manus 中运行

来源

R00T-Kim

R00T-Kim/Terminator

打开 GitHub 仓库查看创作者相关仓库

下载

在 Manus 中运行

CTF Challenge Pipeline

CRITICAL RULES (NEVER VIOLATE)

Local flag file = FAKE — only remote(host, port) has the real flag. Never read local flag and declare FLAG_FOUND
MUST use Agent Teams — never solve by running r2/gdb/python directly (trivial exception only)
Agent model MANDATORY — reverser=sonnet, chain=opus, critic=opus, verifier=sonnet, solver=opus

Pre-checks (auto-executed)

Challenge info: !if [ -n "$1" ] && [ -e "$1" ]; then file "$1" 2>/dev/null && checksec --file="$1" 2>/dev/null | head -5; elif [ -d "$ARGUMENTS" ] 2>/dev/null; then ls -la "$ARGUMENTS" 2>/dev/null | head -10; fi

Knowledge DB prior attempts: !python3 /home/rootk1m/01_CYAI_Lab/01_Projects/Terminator/tools/knowledge_indexer.py search "$ARGUMENTS" 2>/dev/null | head -10 || echo "search unavailable"

Pipeline Rules

MUST solve with Agent Teams. Do NOT solve directly.

Step 0: Pre-analysis (Orchestrator direct)

Read knowledge/index.md — check if already solved or attempted
Verify binary execution: echo "test" | ./<binary> 2>&1
Cannot execute → install libraries first (sudo apt install libc6:i386 etc.)
Basic analysis: file, checksec, strings | head -20

Step 1: Difficulty Assessment → Pipeline Selection

if difficulty == "trivial" (source available, logic bug, one-liner exploit):
    ctf-solver 1-agent (model=sonnet) → reporter
elif type == "pwn" and vuln_clear:
    reverser → chain → critic → verifier → reporter  (5-agent)
elif type == "pwn" and vuln_unclear:
    reverser → trigger → chain → critic → verifier → reporter  (6-agent)
elif type == "reversing" or "crypto":
    reverser → solver → critic → verifier → reporter  (4-agent)
elif type == "web":
    scout → analyst → exploiter → reporter  (4-agent)

Step 2: TeamCreate + Sequential Pipeline

TeamCreate("ctf-<challenge_name>")

All agents: subagent_type="<role>", mode="bypassPermissions", model MUST be specified.

Step 3: Result Collection

FLAG_FOUND → Orchestrator runs solve.py directly to verify
Update knowledge/index.md
TeamDelete cleanup

Failure Protocol

2 failures → Dual-Approach (two different strategies in parallel)
4 failures → WebSearch for writeups

REMINDER: Local flag = FAKE. Always verify via remote(host, port). Agent model parameter is mandatory.

同仓库更多 Skills

同仓库

bounty

R00T-Kim/Terminator

Start Bug Bounty target analysis pipeline. Auto-matches "bounty", "target analysis", "find vulns", "bug hunting", "Immunefi", "Bugcrowd", "H1"

2026-04-226

checkpoint-validate

R00T-Kim/Terminator

Validate agent checkpoint.json. Detect idle, fake completion, error recovery. Matches "checkpoint", "idle", "agent status"

2026-04-226

scout

R00T-Kim/Terminator

Use when firmware binary analysis, attack surface mapping, or endpoint discovery is needed for a target

2026-04-226

coverage-gate

R00T-Kim/Terminator

Check endpoint_map.md coverage at Phase 1→2 transition. Blocks Phase 2 if below 80%. Matches "coverage check", "endpoint coverage", "Phase 2 gate"

2026-03-106

oos-check

R00T-Kim/Terminator

Pre-check if a Bug Bounty finding is Out-of-Scope. Auto-triggered at Phase 0 (full scan) and Phase 1 (per-finding). Matches "OOS", "out of scope", "exclusion check"

2026-03-106

poc-tier

R00T-Kim/Terminator

Classify PoC file quality as Tier 1-4. Run after exploiter completes, before triager-sim. Matches "PoC tier", "PoC quality", "poc quality", "exploit verify"

2026-03-106

name	ctf
description	Start CTF challenge solving pipeline. Auto-matches "ctf", "solve challenge", "pwn", "reversing", "crypto challenge", "wargame"

CTF Challenge Pipeline

CRITICAL RULES (NEVER VIOLATE)

Local flag file = FAKE — only remote(host, port) has the real flag. Never read local flag and declare FLAG_FOUND
MUST use Agent Teams — never solve by running r2/gdb/python directly (trivial exception only)
Agent model MANDATORY — reverser=sonnet, chain=opus, critic=opus, verifier=sonnet, solver=opus

Pre-checks (auto-executed)

Knowledge DB prior attempts: !python3 /home/rootk1m/01_CYAI_Lab/01_Projects/Terminator/tools/knowledge_indexer.py search "$ARGUMENTS" 2>/dev/null | head -10 || echo "search unavailable"

Pipeline Rules

MUST solve with Agent Teams. Do NOT solve directly.

Step 0: Pre-analysis (Orchestrator direct)

Read knowledge/index.md — check if already solved or attempted
Verify binary execution: echo "test" | ./<binary> 2>&1
Cannot execute → install libraries first (sudo apt install libc6:i386 etc.)
Basic analysis: file, checksec, strings | head -20

Step 1: Difficulty Assessment → Pipeline Selection

if difficulty == "trivial" (source available, logic bug, one-liner exploit):
    ctf-solver 1-agent (model=sonnet) → reporter
elif type == "pwn" and vuln_clear:
    reverser → chain → critic → verifier → reporter  (5-agent)
elif type == "pwn" and vuln_unclear:
    reverser → trigger → chain → critic → verifier → reporter  (6-agent)
elif type == "reversing" or "crypto":
    reverser → solver → critic → verifier → reporter  (4-agent)
elif type == "web":
    scout → analyst → exploiter → reporter  (4-agent)

Step 2: TeamCreate + Sequential Pipeline

TeamCreate("ctf-<challenge_name>")

All agents: subagent_type="<role>", mode="bypassPermissions", model MUST be specified.

Step 3: Result Collection

FLAG_FOUND → Orchestrator runs solve.py directly to verify
Update knowledge/index.md
TeamDelete cleanup

Failure Protocol

2 failures → Dual-Approach (two different strategies in parallel)
4 failures → WebSearch for writeups

REMINDER: Local flag = FAKE. Always verify via remote(host, port). Agent model parameter is mandatory.