Agent skill

hub-context

Use when routing user prompts to the right skill or agent. Senses git state, detects platform, classifies intent, and applies context boosts before dispatching.

Stars 163
Forks 31

Install this agent skill to your Project

npx add-skill https://github.com/majiayu000/claude-skill-registry/tree/main/skills/other/other/hub-context

Metadata

Additional technical details for this skill

keywords
context hub routing git-state platform-detection error-signals
triggers
[]
platforms
[
    "all"
]
connections
{
    "enhances": [
        "skill-discovery"
    ]
}
agent affinity
[
    "epost-orchestrator"
]

SKILL.md

Hub Context Skill

Purpose

Defines a context sensing protocol for the /epost smart hub. Before routing, the hub gathers a snapshot of the current working state to make informed routing decisions.

Context Snapshot Protocol

On every /epost invocation, gather these signals before parsing intent:

1. Git State

Run these commands and capture output:

Signal Command Purpose
Branch git branch --show-current Detect feature branch vs main
Staged files git diff --cached --name-only Know what's ready to commit
Unstaged changes git diff --name-only Know what's being worked on
Untracked files git ls-files --others --exclude-standard | head -20 New files not yet tracked
Recent commits git log --oneline -3 Understand recent activity
Merge conflicts git diff --name-only --diff-filter=U Detect blocked state

2. Platform Detection

Determine the dominant platform from changed files:

Extension Pattern Platform Confidence
.tsx, .ts, .jsx, .js, .scss, .css web high
.swift ios high
.kt, .kts android high
.java backend high
.md, .json, .yml unknown low — check directory path

Rules:

  • If 80%+ of changed files share a platform → that platform is dominant
  • If mixed → set platform to multi (triggers orchestrator delegation)
  • If no changed files → check current working directory for platform signals
  • Platform prefix in $ARGUMENTS always overrides auto-detection

3. Error Signals

Check for recent errors (only if platform detected):

Platform Check Command
web TypeScript errors npx tsc --noEmit 2>&1 | head -5 (only if tsconfig.json exists)
web Lint errors Check for recent lint output in terminal
ios Build errors Check for recent xcodebuild failure output
android Build errors Check for recent gradle failure output
any Test failures Check for recent test runner output with "FAIL" or "ERROR"

Important: Error signal gathering should be fast (< 2 seconds). Skip checks that would be slow. TypeScript check is the most valuable — run it if web context is detected.

4. Session Hints

Signal How to Check Purpose
Active plans ls ./plans/*.md 2>/dev/null Detect in-progress work
Data store ls .epost-data/ 2>/dev/null Detect persistent agent state
Branch pattern Parse branch name for feature/, fix/, hotfix/ Infer work type

Context Summary Format

After gathering, produce a structured summary for the routing engine:

[Context] branch: {branch_name} | platform: {platform}
[Context] staged: {count} files ({extensions}) | unstaged: {count} files ({extensions})
[Context] errors: {error_summary} (or "none detected")
[Context] plan: {plan_file} ({status}) (or "no active plan")
[Context] hints: {branch_type}, {other_signals}

Context Boost Table

The routing engine uses context to boost intent categories:

Context Signal Boosts Category Example Routing Effect
TypeScript/build errors detected Fix "what's wrong?" → /fix (auto-detects types)
Staged files present Git or Review "I'm done" → /git-commit
Active plan file exists Build "continue" → /cook with plan context
Test failures detected Fix or Test "help" → /fix (auto-detects test failures)
Feature branch, no changes Build or Plan "what's next?" → continue from plan
Clean main branch Plan or Explore show contextual menu
Merge conflicts detected Fix auto-suggest conflict resolution

Smart Routing

Prompt Classification

  • Dev task (action verbs: cook, fix, plan, test, debug, etc.) → route via intent map
  • Kit question ("which agent", "list commands", "our conventions") → epost-orchestrator
  • External tech question ("how does React...", "what is gRPC") → epost-researcher
  • Conversational (greetings, opinions, clarifications) → respond directly, no routing

Intent → Skill Map

Intent Signal Words Routes To
Build cook, implement, build, create, add, make, continue /cook skill
Fix fix, broken, error, crash, failing, what's wrong /fix skill
Plan plan, design, architect, spec, roadmap /plan skill
Test test, coverage, validate, verify /test skill
Debug debug, trace, inspect, diagnose /debug skill
Review review, check code, audit /review-code skill
Git commit, push, pr, merge, done, ship /git-commit, /git-push, /git-pr
Docs docs, document, write docs /docs-init or /docs-update
Scaffold bootstrap, init, scaffold, new project, new module /bootstrap skill
Convert convert, prototype, migrate /convert skill
A11y a11y, accessibility, wcag /fix-a11y or /review-a11y
Onboard get started, begin, onboard, new to project, what is this /get-started

Routing Rules

  • If user types a slash command explicitly → execute it directly, skip routing
  • If ambiguous → use context boost to break tie; if still ambiguous → ask user (max 1 question)
  • If multi-intent ("plan and build X") → delegate to epost-orchestrator
  • Platform detection delegates to skill-discovery — never hardcode platform logic here

Rules

  • Context gathering MUST complete before intent parsing
  • If a git command fails (not a git repo), skip git state gracefully
  • Never run destructive commands during context gathering
  • TypeScript check should timeout after 5 seconds — skip if slow
  • Context snapshot is read-only — never modify working state

Expand your agent's capabilities with these related and highly-rated skills.

Didn't find tool you were looking for?

Be as detailed as possible for better results