jxucoder

Delegates tasks to Manus when local capabilities are insufficient. Use when tasks require real browser automation, web scraping, persistent compute, long-running autonomous execution, or multi-step workflows with deliverables.

Identifies missing capabilities that warrant new skills. Analyzes repeated friction patterns, failed tasks, and user workarounds to recommend skill creation. Use when discovering-skills finds nothing suitable.

Finds existing skills before building new ones. Searches local installations, official Anthropic repository, and community collections. Evaluates fit and recommends installation or adaptation.

Audits and optimizes context, memory, and skill use for projects. Reviews CLAUDE.md files, .claude/rules/, and skill configurations for efficiency. Use when context feels bloated, skills aren't triggering correctly, or memory needs organization.

Modifies existing skills for new contexts. Forks skills from repositories, adjusts for local conventions, and customizes behavior while preserving core functionality. Use when a skill is close but not quite right.

Extract reusable skills from the current session. Use after solving a non-trivial problem to compound learning across sessions.

Cost-reduction discipline for AI coding sessions. Activates behavioral rules that minimize turns and tokens. Use when: /frugal, save money, reduce cost, efficient session.

Store and manage coding context in git. Use when saving decisions, tracking tasks, planning complex work, or coordinating with other agents. Activates on mentions of "save context", "track task", "planning", or multi-agent coordination.

Map files in a subsystem, trace behavior end-to-end, or run a test and summarize — delegates to cheap Composer scouts so raw output stays out of main agent context. Triggers on 3+ greps, multi-file traces, noisy command output.