| name | reverse-skill |
| description | Use when adapting external reverse-engineering or security workflow packs into OMK, routing APK/binary/JS/browser/API/CTF/report tasks to the right skill, or creating project-local OMK skills from source markdown. |
Reverse Skill
Use this skill to route a reverse-engineering/security workflow before choosing tools, or to turn an external playbook into an OMK Agent Skill.
Core workflow
- Route first: call
reverse_skill_route with the user task. Include targetType, intent, and toolchain when known.
- Read the returned primary skill path before specialized tooling.
- Use returned MCP and hook hints as the lane grant: usually
filesystem, github, playwright/chrome-devtools, plus pre-shell-guard, protect-secrets, and stop-verify.
- Check tool availability only when execution needs a local tool: call
reverse_skill_route with includeToolStatus: true or inspect the command directly.
- When a reusable workflow is missing, call
reverse_skill_create or reverse_skill_from_source to write a project-local skill under .omk/skills/<name>/SKILL.md.
- Verify the generated skill has valid frontmatter, clear triggers, workflow steps, acceptance criteria, and no secrets.
Tools registered by the TS module
reverse_skill_route: target + intent + toolchain scoring across built-in routes.
reverse_skill_create: deterministic skill generator from explicit workflow inputs.
reverse_skill_from_source: markdown-to-skill adapter for external packs.
/reverse-skill <task>: interactive command that records a route decision in the session.
Implementation files
- TS module:
packages/agent/src/harness/reverse-skill.ts
- Project extension:
.omk/extensions/reverse-skill.ts
- Reference route summary:
references/route-map.md
Acceptance
- Output names the primary route, confidence, required tools, MCP hints, hooks, first actions, and blockers.
- Generated skills stay inside the project and follow Agent Skills frontmatter rules.
- Evidence is file/path based; do not store credentials, private keys, raw tokens, or unredacted personal data.