Reverse Engineering Toolkit — a switchable DSH skill plugin
A DeepSeek Harness bundle that registers one model-invocable skill,
reverse-engineering, in the profile-global skill layer. Disable the plugin row
(or the whole bundle) in Plugin Manager and the skill disappears from the session
catalog; enable it again and it comes back.
Repository: https://github.com/MLMLZJ/dsh-reverse-skill
Index: https://github.com/MLMLZJ/dsh-plugins
Companion plugin: https://github.com/MLMLZJ/dsh-install-profiler (installation
profiling, dynamic probing, and the handoff brief that feeds this skill).
Inspired by the viral "REA — Reverse Engineer Anything" project
(https://github.com/morluto/rea), which popularized agent-driven reverse
engineering after adding Android/APK analysis in October 2026. This plugin is an
independent implementation: it ships instructions plus a local toolchain probe,
not REA's engine or MCP server.
What it does
- Adds a
skill catalog entry that the agent loads on demand, then follows to
classify an artifact, probe installed decompilers, work an authorized target,
and report evidence-backed findings.
- Bundles
scripts/probe-tools.ps1 (32 tools across Android/Java/.NET/native/
support, PATH + common install dirs + Android SDK, ~1.5 s, read-only).
- Enforces a blocking authorization gate and an explicit refusal list before any
tool call, and degrades honestly when a decompiler is missing.
Install (once)
git clone https://github.com/MLMLZJ/dsh-reverse-skill.git C:\Tools\dsh-reverse-skill
Then, from a DSH session with Full access or approval:
plugin_manager action=install_bundle target=C:\Tools\dsh-reverse-skill
The install is a link: keep that directory in place, or the plugin breaks.
To update later: git pull in that directory, then toggle the row off and on
(the SKILL.md body is re-read on activation).
Then confirm the row is active with plugin_manager list_plugins (look for
reverse-skill) or approval-free cordis_inspect_query on the Config
provider filtered by @local/dsh-reverse-skill.
On / off
plugin_manager action=set_plugin target=reverse-skill enabled=false — unregisters
the skill; the next turn receives a replacement catalog without it.enabled=true — re-registers it (the SKILL.md body is re-read on activation,
so content edits take effect after a toggle, without restarting DSH).action=remove_bundle target=@local/dsh-reverse-skill — uninstalls completely.
Layout
package.json bundle manifest (dsh.bundle.patch)
cordis.patch.yml inserts row id `reverse-skill`
index.js registers the skill via ctx.skills.register()
locale/en.json, zh.json Plugin Manager title/description
icon.svg original icon
skills/reverse-engineering/
SKILL.md the instructions the model loads
references/tool-recipes.md
scripts/probe-tools.ps1 read-only toolchain probe
Safety
The skill refuses DRM/licence/anti-cheat bypass, credential theft, unauthorized
targets, live-service interaction, and redistribution of decompiled material. It
never executes a target on the host; dynamic analysis is limited to a disposable
device/VM the user authorizes. Decompiling a client never reveals server-side
logic — the skill states that limit instead of overselling.
