malskill

Drive the MCPwn Kali-backed MCP server cleanly and fast: create a session, discover tools via the catalog (list_catalog/get_tools/get_tool/run_tool) instead of guessing names, pick the right execution path (execute_command vs detach+poll_job vs interactive shell), move files with the correct mechanism (write_workspace_file, request_upload/request_download data plane, import_artifact_to_workspace, upload_to_target, list_payloads/get_payload), and bring up connectivity (tunnel_up VPN/proxy/forward, tunnel_revshell, penelope reverse shells, run_in_shell/stabilize_shell). Use whenever running MCPwn / mcpwn_* tools, a Kali MCP, or when a run stalls on timeouts, lost output, missing routes, wrong tool names, or clumsy file transfer.

Keep the context window lean when reading large data. Use before opening big files, logs, dumps, PCAPs, decompiler output, research corpora, or a folder of worker artifacts. Enforces grep-first + windowed reads, extract-don't-hoard, and context quarantine (delegate a huge read to a sub-agent that returns a digest).

Auth assessment: web impact-validation; SQLi, SSTI, XXE, command injection, SSRF, XSS, uploads, deserialization, smuggling, WAF/parser checks.

Mode: /1337 - structured operator behaviour for coding and security; forces explicit reasoning, fast decisions, todos/lists, exact terms, evidence, verification, safety override.

Lab/CTF: pwn/binary challenges; native binaries, memory corruption, format strings, heap/ROP/SROP, shellcode artifacts, seccomp, kernel labs.

Auth/lab: reverse engineering methodology; malware triage, patch diffing, firmware/protocol RE, protections, exploitability handoff evidence.

Mode: /1337-brain - maintain an Obsidian vault as an AI second brain for project knowledge. Use to ingest sources, create atomic linked wiki notes, build maps of content (MOCs), answer grounded questions through a hot-cache, index, then page retrieval ladder, dedupe, audit for contradictions and dead links, keep a recent-context cache, and log changes. Prefers the Obsidian MCP (list/read/search/create/update notes); falls back to the filesystem. Assumes Obsidian and the Obsidian MCP are installed. Trigger on /1337-brain or subcommands init|ingest|ask|link|moc|dedupe|audit|project|profile|log|hot, or when the user wants to capture, organize, distill, or retrieve durable project knowledge. Markdown is the source of truth; never invent facts; preserve source paths; keep links selective.

Investigation discipline for extremely complex problems where the root cause, exploit path, or solution is unknown: hard bugs, flaky systems, CTF challenges, reverse engineering puzzles, incident triage, multi-system failures, and research questions backed by data. Forces explicit hypothesis generation, falsifiable predictions, prioritized experiments, and evidence-based iteration instead of trial-and-error patching or confirmation-biased reasoning. Use when symptoms are far from causes, when guesses keep failing, or when a problem spans tools, layers, or unknowns that defeat linear debugging.

Retry discipline for stuck work. Use when an approach fails repeatedly, when grinding a side problem (env setup, missing tool, credentials, file transfer) instead of the goal, or when a debug/exploit/build attempt is not converging. Enforces evidence-first pivots, a 3-strikes rule, and honest BLOCKED reporting over thrash - while resisting premature give-up: pivot the dead path, but hold the objective while budget and untried approaches remain.

Auth assessment: post-compromise host methodology; shell state, local privesc triage, secret-exposure review, pivot planning, evidence.

Container methodology: Identifying containerization limits, Docker/K8s misconfigurations, and executing escapes to the host node.

Lab/CTF: satellite and space-system challenges; CCSDS Space Packets, TC/TM/AOS/USLP Space Data Link transfer frames, CLTU/BCH/Reed-Solomon coding, COP-1/FARM sequence control, ECSS PUS services, SDLS/SDLS-EP link-layer crypto and CryptoLib CVEs, CubeSat Space Protocol (CSP), TLE/SGP4 orbits, GPS L1 GNSS spoofing with gps-sdr-sim/HackRF, and ground-segment targets (Yamcs, OpenC3 COSMOS, NASA cFS, NOS3, Kubos) including known 2025–2026 RCE/path-traversal/DoS CVEs. Use when a task mentions spacecraft ID, APID, virtual channel, telecommand/telemetry, transfer frame, ground station, modem, ASM 1ACFFC1D, SPI/OTAR/ARSN, or a downlink/uplink byte stream to build or decode.

Lab/CTF: general challenge-solving methodology across every category; artifact triage, spec/protocol identification, oracle-driven iteration from target error messages, minimal-matrix brute forcing, harness building with pwntools, parallelizing independent challenges/variants with subagents, and safe flag extraction, verification, and submission. Use to plan or drive any flag-style challenge, to route to the right domain *-ctf skill, or whenever a remote service replies with parseable errors that can steer the next input.

Auth/lab dev: Windows detection-resilience research; syscall dispatch, stack traces, sleep-state, memory permissions, ETW/AMSI telemetry tradeoffs.

Lab/CTF: web challenges; HTTP apps, APIs, browser clients, auth, uploads, SSRF, XSS, SQLi, SSTI, XXE, deserialization, smuggling.

Auth assessment: AD decision support; LDAP/SMB/Kerberos, BloodHound graph, ADCS, NTLM relay risk, ACL/trust paths, evidence-first routing.

Machine-oracle verification for security findings: promote a finding to VERIFIED only when a named oracle re-runs the exploit N/N against the live target AND a negative control fails on a safe surface, then package a portable proof capsule the client can replay. Use before reporting web/API/auth findings, when triaging scanner output (nuclei/nikto/zap/sqlmap), when an LLM or sub-agent asserts a vulnerability without proof, or when a report must separate proven findings from candidates. Tool-agnostic; pairs with web-exploit, vuln-exploit, and report-generation techniques.

Auth/lab ref: Python-based AD Certificate Services testing tool.

Binary reverse engineering with Python: analyze, parse, and disassemble ELF/PE executables using pwntools, capstone, Frida, and custom parsing tools. Use when understanding malware, debugging binary failures, analyzing section structure, extracting strings/entropy, or instrumenting runtime behavior.

Auth/lab ASM patterns; x86-64/ARM64, syscalls, SSN resolution, stack traces, PEB/IAT-free lookup, PIC data access, ETW/AMSI telemetry, BOF/loader review.

Flipper Zero app development (FAP), Sub-GHz scripting, badUSB automation, and GPIO hardware bridging.

Auth/lab ref: Saleae Logic 2 capture and export workflow: `.sal` traces, raw CSV export, analyzer-table export, and automation API control.

Auth/lab ref: Python CTF/exploitation framework for interacting with remote services, local processes, and binary exploitation.

Create, update, or refactor repository-root and nested AGENTS.md files for AI coding agents. Use when the user asks to bootstrap AGENTS.md, replace tool-specific instruction files with a shared open format, compress overly verbose agent instructions, document build/test commands for agents, or design minimal project instructions for monorepos and subprojects.

Design, build, and tune Claude agents — primarily Claude Code subagents (the Markdown + YAML files under .claude/agents/ or ~/.claude/agents/ that Claude delegates to), and also Claude Managed Agents (the cloud agent harness on the API). Use when asked to create a Claude agent or subagent, write agent frontmatter (name, description, tools, model, skills, permissionMode, hooks, memory), design a focused reviewer/debugger/researcher agent, scope tool access, route work to a cheaper model, build a team of delegating agents, or fix an agent that never triggers or sees the wrong context. Not for AGENTS.md / CLAUDE.md instruction files (use agent-md-creator), portable Agent Skills (use skill-creator), or OpenCode agents (use opencode-agent-creator).

Enumerate CVEs for a vulnerability class, find credible public PoC references with GitHub preferred, verify collected URLs, and write structured Markdown evidence for downstream use. Use when asked to build or refresh a CVE list, PoC tracker, exploit reference sheet, or vulnerability-type digest from public sources.

File-backed deep research with recursive link-following, multi-tool web fetching, and step-by-step synthesis. Use when the user asks to research, investigate, analyze, or summarize a topic in depth; when a thorough answer requires gathering and cross-referencing multiple sources; or when output must be comprehensive, cited, and not limited by context window size. For CVE/exploit/threat-intel research → use deep-research-offensive.

File-backed offensive security research with recursive link-following, multi-tool fetching (Jina Reader, Tavily, Playwright), and step-by-step synthesis. Use when researching CVEs, vulnerabilities, exploits, attack chains, PoC code, OSINT targets, red team planning, or threat intelligence. Saves each useful page to intermediate files, follows linked sources recursively, and produces a comprehensive research document not limited by context size.

Triage external technical feedback before applying it. Use for code reviews, scanner findings, exploit PoC notes, blog advice, LLM suggestions, issue comments, and advisory recommendations. Verifies context fit, evidence, risk, and minimal changes instead of accepting or rejecting feedback performatively.

Use after an approved design, spec, issue, or requirement set and before multi-step implementation. Produces bite-sized tasks with exact files, commands, verification gates, review checkpoints, and stop conditions for coding, skill curation, offensive tooling, or research automation.

Targeted post-triage online research for a known technical problem signature, not broad discovery. Use after the agent has already analyzed the artifact, ranked hypotheses, and hit a wall, to find the missing hint in papers, blogs, articles, public writeups, source discussions, specifications, commits, issues, changelogs, advisories, PoCs, or implementation notes. Useful for cryptography, protocol debugging, reversing, AI/ML behavior, web/API behavior, exploit constraints, version-specific bugs, build/runtime errors, and standards mismatches where one external clue unlocks the next local test.

Design, build, and tune OpenCode CLI agents and multi-agent teams. Use when asked to create an OpenCode agent, a subagent, an orchestrator/supervisor, a roster of specialists, a swarm/parallel fan-out, or an agent .md / opencode.json config. Covers agent modes (primary/subagent/all), the built-in agents (build, plan, general, explore, scout), the native Task tool for hierarchical delegation, permission.task dispatch whitelists, per-agent model/cost routing (and the no-per-task-model constraint plus the background-agents plugin workaround), the supervisor/swarm/mesh/template topologies and which are natively supported, custom tools, plugins, and secrets. Not for AGENTS.md instruction files (use agent-md-creator), generic Agent Skills (use skill-creator), or Claude Code subagents (use agents-claude-creator).

Build, structure, test, and ship OpenCode CLI plugins — the JS/TS modules that extend OpenCode through the @opencode-ai/plugin API. Use when asked to create an OpenCode plugin, add a plugin hook (event, config, chat.message, chat.params, chat.headers, permission.ask, tool.execute.before/after, shell.env, command.execute.before, tool.definition, auth, provider, dispose, or any experimental.* hook for system-prompt/messages transform or compaction), register a custom tool() backed by a zod schema, scaffold a plugin package.json + tsconfig, wire the `plugin` array in opencode.json, install a local plugin via shim, or publish a plugin to npm. Covers the PluginInput context (client, $, directory, worktree, project, serverUrl), the Hooks return shape, ToolContext/ToolResult, the Bun runtime, and cross-platform gotchas. NOT for OpenCode agents/subagents (use opencode-agent-creator), AGENTS.md files (use agent-md-creator), or portable Agent Skills (use skill-creator).

Create, review, package, and troubleshoot Pi coding-agent extensions and Pi packages. Use when asked to build TypeScript extensions for Pi, register tools with pi.registerTool, subscribe to pi.on lifecycle/tool/input/session events, add slash commands, custom UI/widgets/renderers, package resources through package.json pi.extensions/skills/prompts/themes, define custom subagent markdown agents/chains for pi-subagents, install via pi -e or pi install, or adapt examples such as pi-fork, pi-minimal-subagent, Hypa pi-hypa, permission gates, protected paths, subagents, command rewriters, custom providers, and dynamic resources.

Safely evaluate, adapt, and rewrite raw public proof-of-concepts into reliable, offline-capable exploits free of backdoors.

Create, update, rewrite, or shorten repository README.md files so they stay concise, well-sectioned, non-redundant, and written in clear English. Use when the user asks to create a new README, improve or refresh an existing one, standardize structure and tone, remove bloat, or add practical setup and usage guidance, selective badges or callouts, and small ASCII architecture diagrams only when the architecture or flow is important to understanding the project.

Design, create, update, and package Agent Skills following the open AgentSkills specification (agentskills.io). Use when asked to create a new skill, improve an existing skill, scaffold a skill directory, validate a SKILL.md, or package a skill into a distributable .skill file.

Lab/CTF: AI/ML challenges; model artifacts, checkpoints, embeddings, LoRA, classifiers, model APIs, RAG/tool-use, adversarial ML.

Lab/CTF: blockchain/Web3 challenges; Solidity/Vyper/EVM, ABI, storage, calldata, proxies, delegatecall, tx traces, Foundry/Hardhat.

Lab/CTF: cloud security challenges; AWS/GCP/Azure creds, buckets, IAM, metadata, KMS/secrets, snapshots, object versions, identity chains.