CTF Automation — one-shot triage and category setup

All scripts live in this directory. Every script emits JSON to stdout (when --json is given) so you can pipe into jq or the next stage of a chain.

Entry point: triage

bash /home/ubuntu/.claude/skills/ctf-automation/triage.sh <challenge-dir>

Outputs:

1. <challenge-dir>/.ctf-triage.json — machine-readable fingerprint
2. <challenge-dir>/.ctf-triage.md — markdown report with *pointers into `ctf-/SKILL.md#pattern-recognition-index`**

After triage, pick the indicated category script:

Tool inventory philosophy

Each script performs a command -v check for every external tool it uses. Missing tools do not crash — they print the exact apt install … / go install … / pip install … command for the missing binary and continue on what remains available.

Scripts in this directory

• triage.sh — master triage; emits JSON + markdown pointing to Pattern Recognition Index sections
• pwnsetup.sh — checksec → detect libc → libc.rip lookup → patchelf → generate exploit.py pwntools template pre-wired for local+remote
• cryptosetup.py — parses challenge files; detects RSA (n,e,c), ECDSA sigs (r,s), lattice shapes, post-quantum params (Kyber/Dilithium/Falcon); generates a Sage script stub with correct imports
• websetup.sh — chained recon: subfinder → httpx → katana → ffuf → nuclei, merges to single JSON
• foreniq.sh — RF/audio/logic-analyzer pipeline: GQRX UDP / file → sox 22050Hz mono → multimon-ng -a POCSAG512/1200/2400 -f alpha; .sr files → pulseview CLI export
• aiprobe.py — LLM endpoint auto-attack: argument injection on tool-allow-lists, DNS rebind, language-guardrail-gap, metadata exfil, reverse-order prompt, literal-policy flip. Emits finding JSON per attack.
• ctfd.py — CTFd platform client: sync challenges + download files, run triage on workspace, submit flags, show progress, pick next challenge.

CTFd workflow (ctfd.py)

Full competition workflow against any CTFd instance:

SKILLS=~/.claude/skills/ctf-automation

# 1. init workspace (once per CTF)
python3 $SKILLS/ctfd.py init \
  --url https://ctf.example.com \
  --token <your-api-token> \
  --out ./workspace

# 2. download all challenges + files
python3 $SKILLS/ctfd.py sync --dir ./workspace

# 3. fingerprint everything
python3 $SKILLS/ctfd.py triage --dir ./workspace

# 4. pick next target (lowest points first)
python3 $SKILLS/ctfd.py next --dir ./workspace
# → prints: Dir, connection info, triage hint

# 5. work the challenge … find flag …

# 6. submit
python3 $SKILLS/ctfd.py submit "CTF{...}" --chal challenge-name --dir ./workspace

# 7. loop back to step 4
python3 $SKILLS/ctfd.py status --dir ./workspace

Workspace layout written by sync:

workspace/
  .ctfd.json          ← config + challenge index
  pwn/
    heap-overflow/
      vuln  libc.so.6 Dockerfile
      .meta.json       ← id, pts, description, connection_info
      .triage.json     ← from triage step
      .solved          ← written on correct submit
  web/
    login-bypass/
      ...

next sorts unsolved challenges by points (easiest first); filter by category with --category pwn.

Pattern Recognition dispatch

triage.sh reads the file list and dependency manifests, then writes pointers like:

ctf-pwn/SKILL.md#pattern-recognition-index → row "MAP_FIXED exposed"
ctf-crypto/SKILL.md#pattern-recognition-index → row "post-quantum KEM"
ctf-misc/ai-ml.md → section "Argument injection on allow-listed tools"

The calling agent (/solve-challenge) reads the markdown report and dispatches to the skill(s) indicated. This replaces guessing the category from the user prompt — we dispatch on what is actually in the challenge folder.

Chain example

DIR=/tmp/ch-42
bash triage.sh "$DIR" --json | jq '.hints[]'
# If hints mention "libc":
bash pwnsetup.sh "$DIR/vuln"
# If hints mention "ai_endpoint":
python3 aiprobe.py http://chal.example/api --json > findings.json

Exit codes

• 0 — triage ran, report written
• 2 — directory does not exist / unreadable
• 3 — no recognisable artefacts (empty, or only text README)

Never exit 1; that's reserved for unexpected script errors, which indicate a bug to fix.