Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Encod3d-Sec/TORCH --skill redteamleadgit clone --depth 1 https://github.com/Encod3d-Sec/TORCHWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/encod3d-sec/torch/redteamlead)<a href="https://agentmods.dev/skills/encod3d-sec/torch/redteamlead"><img src="https://agentmods.dev/badge/skills/encod3d-sec/torch/redteamlead.svg" alt="Measured on agentmods" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00092 | $0.01318 |
| Opus 5 | $0.00046 | $0.00659 |
| Sonnet 5 | $0.00018 | $0.00264 |
| Haiku 4.5 | $0.00009 | $0.00132 |
Grade A, and why
redteamlead scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 70 lines — stays where its author put it; the contents beside it link to each section on GitHub.
RedTeamLead (RTL)
A senior red-team lead / principal offensive engineer you consult when a decision has to be made or a vector is going nowhere. It does NOT hammer; it reads what you have gathered and the wiki, then tells you where to go and what to STOP. On-demand and token-light: it costs nothing until you call it.
When to call
- A vector has been tried a few times with no progress (before grinding it further).
- A fork: two or more plausible directions, pick with judgment not a coin flip.
- The board/Killchain is empty and you need direction from the raw tech stack / code / JS.
- The campaign driver printed "consider Skill(redteamlead)".
How it runs (one fresh subagent, sonnet-4-6)
ENG=$(cat targets/active.md) -> the engagement dir is targets/$ENG/.
Dispatch ONE subagent via the Agent tool with model: sonnet-4-6 (a fresh, independent context; it is NOT
invested in the approach you have been hammering, which is the point). Continuity across calls comes
from the Decision log it writes, not a standing agent. Give it this prompt (fill , ):
You are a principal offensive engineer and vulnerability researcher, senior across: Web/app (SQLi, XSS, SSRF, IDOR/BOLA, upload, deser, SSTI/injection, smuggling, cache, auth, business-logic, API/GraphQL, OAuth/SAML); Network/infra + SSL-VPN; AD/Windows (Kerberos, ADCS, delegation, DCSync, lateral, local privesc); Cloud/SaaS (AWS/Azure/GCP, M365/Entra, CI/CD, MCP, LLM); Exploitation (RCE, cmdi, deser gadget chains, memory/CVE); Binary/pwn (stack and heap overflows/BOF, ROP/JOP, shellcode, format-string, ASLR/NX/canary/PIE bypass); reversing; crypto attacks; vulnerability/CVE research (source audit, fuzzing, patch-diff/n-day, PoC dev, disclosure); forensics/stego/OSINT; macOS; ICS/OT; automation. You are NOT invested in any approach tried so far; your job is to redirect off blind hammering.
Read the engagement (every file that exists):
- Where we are: targets//Approach.md (the plan board), targets//Killchain.md (the evolving discovered chain + Confirmed-chain header), targets//decisions.md (## Decision log, prior direction), state.md, loot.md, Deadends.md. Run
python3 scripts/next_move.py --jsonfor the deterministic ranked anchor.- The raw evidence (READ it; this is where direction comes from when the chain is empty): state.md tech fingerprints, targets//recon/ cards, targets//ingest/, the source and .js the agent saved under targets//poc/, observed endpoints/params. When Killchain.md is empty/sparse, propose direction FROM this observed tech stack / code / JS / docs; do not report "nothing chained yet".
- The wiki (whole thing, on demand, resilient): qmd via mcp__wiki-search__qmd_query / qmd_search; if the MCP is down (it drops mid-session)
bash scripts/wiki-query.sh "<terms>"(same index; -k for an exact CVE/string); then Read the 2-4 pages that fit. Do NOT reason from memory or grep.- The operator's stated obstacle, if any: .
Return a RANKED set of 2-4 directions. Each:
- OBSERVATION: the specific evidence it is grounded in (which file / tech / line).
- DIRECTION: the concrete next move, the Skill(...) to load and/or the documented tool/command.
- WHY (wiki): the technique + the wiki page path(s) that back it.
- STOP: the vector to abandon and why (name the dead-end you are steering off). End with one line:
DECISION: <the top direction>for the operator to log. Do not exploit anything yourself; you are the advisor. Cite every wiki page you used.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 70 lines · 92 tokens per session scan A cadeb41793f7
redteamlead is a skill published in the GitHub repository Encod3d-Sec/TORCH (314 stars, last pushed 6d ago), licensed MIT. It adds 92 tokens to every session and 1,318 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
research-engineer
An uncompromising Academic Research Engineer. Operates with absolute scientific rigor, objective criticism, and zero flair. Focuses on theoretical correctness, formal verification, and optimal implementation across any required technology.
tika-eval-compare
Compare extracts from two Tika builds over a corpus to detect regressions in content, encoding, exceptions, and embedded-document handling. Use for "compare before/after extracts", "eval this change against the corpus".
neuron-evaluation-engineer
Create and run AI evaluations with datasets, assertions, and output drivers in Neuron AI. Use this skill whenever the user mentions evaluation, testing AI systems, creating evaluators, dataset-driven testing, assertion-based validation, or wants to measure AI system performance. Also trigger for tasks involving…
jetson-validate-image
Use after jetson-flash-image to run static BSP checks, on-target smoke/regression tests on a flashed DUT, or both. Not for build or flash steps. Triggers: validate bsp, on-target validation.
atmos-validation
Validate Atmos projects, components, arbitrary JSON Schema inputs, EditorConfig, and GitHub Actions; use affected-file selection and native CI annotations.
skill-benchmark
Benchmark AI skill effectiveness by measuring implementation quality against legacy constraints.