Description
The usual way an AI coding assistant fails at reverse engineering is that it simply does not know which tool to reach for. Hand it an APK and it flails between jadx, apktool and Frida; hand it a binary and it hesitates between IDA and radare2. Encrypted frontend parameters, a CTF challenge, a packet capture — each needs a different playbook, and the agent has neither a shared methodology nor any memory of the mistakes it made last time, so it repeats them.
reverse-skill is a skill router for that situation. It does not write exploits for you; when a client such as Claude Code, Cursor or Cline meets a reverse-engineering or authorized pentest task, it routes the task to the right methodology, checks whether the required tools are present and bootstraps them on demand, then runs a repeatable workflow instead of letting the agent guess commands. The routing core is driven by one structured configuration — 44 rules and 175 regression cases, validated by cross-platform CI on Windows and Ubuntu, kept separate from client adapters.
Its scope is stated plainly: lawful security research, education, CTF competitions, and systems you own or are explicitly authorized to assess. A scope gate is built into the flow — nothing acts against a target until authorization is confirmed. The project prohibits unauthorized access, scanning or exploitation outright and holds the user responsible for staying within the law and the agreed scope.
Automatic task routing: APK, ELF, frontend JS, packet captures, CTF and more are matched by rule to the corresponding skill module and methodology, so the agent no longer guesses where to start.
Authorization gate first: scope and authorization are confirmed before any target work, making "get authorized first" a hard prerequisite in the workflow rather than an afterthought.
On-demand toolchain bootstrapping: once routed to a scenario it checks for the required tools, MCP servers or scripts and pulls in what is missing, instead of pre-scattering tools across machines.
Broad scenario coverage: Android/iOS, binaries, .NET, frontend JS, malware and YARA, firmware/IoT, patch diffing and CTF each have their own skill module and playbook.
Evidence chain and reporting: timelines are recorded as Evidence to Finding to Path, producing reviewable reports and field notes for handoff and retrospectives.
Self-evolving knowledge base: what worked in each case is distilled into a field-journal and reused on the next similar task, cutting repeated mistakes.
Client-neutral: supports Claude Code, Cursor, Cline, Kiro and other AI coding clients, with the routing core separated from client adapters. MIT licensed.
reverse-skill is a skill router for that situation. It does not write exploits for you; when a client such as Claude Code, Cursor or Cline meets a reverse-engineering or authorized pentest task, it routes the task to the right methodology, checks whether the required tools are present and bootstraps them on demand, then runs a repeatable workflow instead of letting the agent guess commands. The routing core is driven by one structured configuration — 44 rules and 175 regression cases, validated by cross-platform CI on Windows and Ubuntu, kept separate from client adapters.
Its scope is stated plainly: lawful security research, education, CTF competitions, and systems you own or are explicitly authorized to assess. A scope gate is built into the flow — nothing acts against a target until authorization is confirmed. The project prohibits unauthorized access, scanning or exploitation outright and holds the user responsible for staying within the law and the agreed scope.
Features
Automatic task routing: APK, ELF, frontend JS, packet captures, CTF and more are matched by rule to the corresponding skill module and methodology, so the agent no longer guesses where to start.
Authorization gate first: scope and authorization are confirmed before any target work, making "get authorized first" a hard prerequisite in the workflow rather than an afterthought.
On-demand toolchain bootstrapping: once routed to a scenario it checks for the required tools, MCP servers or scripts and pulls in what is missing, instead of pre-scattering tools across machines.
Broad scenario coverage: Android/iOS, binaries, .NET, frontend JS, malware and YARA, firmware/IoT, patch diffing and CTF each have their own skill module and playbook.
Evidence chain and reporting: timelines are recorded as Evidence to Finding to Path, producing reviewable reports and field notes for handoff and retrospectives.
Self-evolving knowledge base: what worked in each case is distilled into a field-journal and reused on the next similar task, cutting repeated mistakes.
Client-neutral: supports Claude Code, Cursor, Cline, Kiro and other AI coding clients, with the routing core separated from client adapters. MIT licensed.
