AgentSkillsCN

context-mate

Context Mate 工具集的入口——涵盖与 Claude Code 自然流畅协作的技能、代理与斜杠命令。从项目生命周期(/explore-idea → /plan-project → /wrap-session),到跨上下文窗口的会话交接,再到专精于特定任务的开发者代理,以及贯穿始终的质量审计,一应俱全。 适用于:启动新项目、深入了解工具集、寻求工作流指导时使用。 “一切的关键在于‘上下文’,maaate!”

SKILL.md
--- frontmatter
name: context-mate
description: |
  Entry point to the Context Mate toolkit - skills, agents, and slash commands that work with Claude Code's
  natural flow. Project lifecycle (/explore-idea → /plan-project → /wrap-session), session handoff
  across context windows, developer agents for specialized tasks, and quality auditing.

  Use when: starting new projects, understanding the toolkit, needing workflow guidance.
  "It's all about the context, maaate!"
user-invocable: true

Context Mate

A toolkit that works with Claude Code's natural flow. Use what helps, ignore what doesn't.


When This Skill Activates

When context-mate is invoked, analyze the project first before recommending tools.

Step 1: Quick Project Scan

Check for these files (use Glob, don't read contents yet):

File/PatternIndicates
SESSION.mdSession tracking active
IMPLEMENTATION_PHASES.mdPhased planning in use
PROJECT_BRIEF.mdProject explored/planned
CLAUDE.md or .claude/AI context exists
.claude/rules/Correction rules present
package.json or requirements.txtHas dependencies
tests/ or *.test.*Has test infrastructure

Step 2: Git State (if git repo)

bash
git status --short            # Uncommitted changes?
git log --oneline -3          # Recent commit messages?

Step 3: Assess Stage and Recommend

Project Stages:

StageSignsRecommend
New ProjectNo CLAUDE.md, no phases/explore-idea or /plan-project
Active DevelopmentSESSION.md or phases exist/continue-session, developer agents
Maintenance ModeDocs exist, no SESSION.md/plan-feature for new work, project-health for audits
Mid-SessionUncommitted changes + SESSION.mdContinue current work, /wrap-session when done

Step 4: Brief Output

Tell the user:

  1. What's already set up (e.g., "You have SESSION.md and phases - mid-project")
  2. What would help now (e.g., "Run /continue-session to resume")
  3. What's available but not in use (e.g., "No tests yet - test-runner available")

Example:

Project Analysis

CLAUDE.md - AI context configured ✓ SESSION.md - Session tracking active (Phase 2 in progress) ✓ .claude/rules/ - 3 correction rules ○ No test files detected

Recommendations:

  • Run /continue-session to resume Phase 2 work
  • Use commit-helper agent when ready to commit
  • Consider test-runner agent when adding tests

Keep it under 10 lines. Don't overwhelm - just highlight what's relevant.


The name has a double meaning:

  1. Your friendly context companion (the toolkit)
  2. "It's all about the context, maaate!" (the philosophy)

This isn't "The Correct Way To Do Things" - these tools exist because context windows are real constraints, not because we're dictating methodology.


Quick Reference

Slash Commands (type these)

CommandWhat it does
/context-mateAnalyze project, recommend tools
/explore-ideaStart with a vague idea
/plan-projectPlan a new project
/plan-featurePlan a specific feature
/wrap-sessionEnd work session
/continue-sessionResume from last session
/docs-initCreate project docs
/docs-updateUpdate docs after changes
/briefPreserve context before clearing
/reflectCapture learnings → rules, skills, memory
/releasePrepare for deployment

Agents (Claude uses these automatically)

AgentWhat it does
commit-helperWrites commit messages
code-reviewerReviews code quality
debuggerInvestigates bugs
test-runnerRuns/writes tests
build-verifierChecks dist matches source
documentation-expertCreates/updates docs
orchestratorCoordinates multi-step work

Skills (background knowledge)

SkillWhat it provides
project-planningPhase-based planning templates
project-session-managementSESSION.md patterns
docs-workflowDoc maintenance commands
deep-debugMulti-agent debugging
project-healthAI-readability audits
developer-toolboxThe 7 agents above

The Toolkit at a Glance

code
┌─────────────────────────────────────────────────────────────┐
│                    PROJECT LIFECYCLE                        │
├─────────────────────────────────────────────────────────────┤
│  /explore-idea → /plan-project → [work] → /wrap-session    │
│       ↓              ↓              ↓           ↓          │
│  PROJECT_BRIEF   PHASES.md     SESSION.md   git checkpoint │
│                                     ↓                      │
│                              /continue-session             │
│                                     ↓                      │
│                              [resume work]                 │
│                                     ↓                      │
│                    /reflect → /release                     │
└─────────────────────────────────────────────────────────────┘

When To Use What

You want to...Use this
Explore a vague idea/explore-idea
Plan a new project/plan-project
Plan a specific feature/plan-feature
End a work session/wrap-session
Resume after a break/continue-session
Create/update docs/docs-init, /docs-update
Debug something stubborndeep-debug skill
Review code qualitycode-reviewer agent
Run tests with TDDtest-runner agent
Prepare a git commitcommit-helper agent
Verify build outputbuild-verifier agent
Check docs are AI-readablecontext-auditor agent
Validate workflows workworkflow-validator agent
Check session handoff qualityhandoff-checker agent

Component Skills

Project Lifecycle (project-workflow)

Nine integrated commands for the complete project lifecycle:

CommandPurpose
/explore-ideaBrainstorm and validate project concepts
/plan-projectGenerate phased implementation plan
/plan-featurePlan a specific feature addition
/docs-initCreate initial project documentation
/docs-updateUpdate docs after changes
/wrap-sessionEnd session with git checkpoint
/continue-sessionResume from SESSION.md
/reflectReview progress and plan next steps
/releasePrepare for deployment/release

Invoke: Skill(skill: "project-workflow")

Session Management (project-session-management)

Track progress across context windows using SESSION.md with git checkpoints.

  • Converts IMPLEMENTATION_PHASES.md into actionable tracking
  • Creates semantic git commits as recovery points
  • Documents concrete next actions for resumption
  • Prevents context loss between sessions

Invoke: Skill(skill: "project-session-management")

Developer Agents (developer-toolbox)

Seven specialized agents for common development tasks:

AgentUse For
commit-helperGenerate meaningful commit messages
code-reviewerSecurity, quality, architecture review
debuggerSystematic bug investigation
test-runnerTDD workflows, test creation
build-verifierVerify dist/ matches source
documentation-expertCreate/update project docs
orchestratorCoordinate multi-step projects

Invoke: Skill(skill: "developer-toolbox")

Deep Debugging (deep-debug)

Multi-agent investigation for stubborn bugs that resist normal debugging.

  • Spawns parallel investigation agents
  • Cross-references findings
  • Handles browser/runtime issues
  • Best when going in circles on a bug

Invoke: Skill(skill: "deep-debug")

Quality Auditing (project-health)

Three agents for AI-readability and workflow quality:

AgentPurpose
context-auditorCheck if docs are AI-readable (score 0-100)
workflow-validatorVerify documented processes work (score 0-100)
handoff-checkerValidate session continuity quality (score 0-100)

Invoke: Skill(skill: "project-health")

Documentation Lifecycle (docs-workflow)

Four commands for documentation management:

CommandPurpose
/docsQuick doc lookup
/docs-initCreate initial docs
/docs-updateUpdate after changes
/docs-claudeGenerate AI-optimized CLAUDE.md

Invoke: Skill(skill: "docs-workflow")


Core Concepts

Sessions ≠ Phases

Sessions are context windows (2-4 hours of work before context fills up).

Phases are work units (logical groupings like "Phase 1: Database Setup").

A phase might span multiple sessions. A session might touch multiple phases. They're independent concepts.

Checkpointed Progress

Git commits serve as semantic checkpoints, not just version control:

bash
# Bad: commits as save points
git commit -m "WIP"
git commit -m "more changes"

# Good: commits as progress markers
git commit -m "Complete Phase 1: Database schema and migrations"
git commit -m "Phase 2 partial: Auth middleware working, UI pending"

When resuming via /continue-session, these commits tell the story of where you are.

Progressive Disclosure

Skills load incrementally to preserve context:

  1. Metadata (~50 tokens) - Always in context, triggers skill loading
  2. SKILL.md body (<5k words) - Loaded when skill activates
  3. Bundled resources - Loaded as needed (templates, references, scripts)

This means a 50-skill toolkit only costs ~2,500 tokens until you actually use something.

Skills Teach, Rules Correct

Two complementary knowledge systems:

SkillsRules
Location~/.claude/skills/.claude/rules/ (project)
ContentRich bundlesSingle markdown files
PurposeTeach how to use XCorrect outdated patterns
ExampleHow to set up Tailwind v4Fix v3 syntax Claude might suggest

Rules are project-portable - they travel with the repo so any Claude instance gets the corrections.

Sub-agents for Isolation

Heavy tasks (code review, debugging, testing) run in sub-agents to:

  • Keep verbose output out of main context
  • Allow parallel execution
  • Provide specialized tool access
  • Return concise summaries

Getting Started

New Project

code
/explore-idea    # Optional: clarify what you're building
/plan-project    # Generate phased plan
                 # Work on Phase 1...
/wrap-session    # End with checkpoint

Resuming Work

code
/continue-session    # Reads SESSION.md, suggests next steps
                     # Continue working...
/wrap-session        # Checkpoint again

Adding a Feature

code
/plan-feature    # Plan the specific feature
                 # Implement...
/wrap-session    # Checkpoint

Debugging Session

code
# If normal debugging isn't working:
Skill(skill: "deep-debug")
# Spawns investigation agents

The Philosophy

Context windows are real. They fill up. Work gets lost. Sessions end.

These tools don't fight that - they work with it:

  • SESSION.md captures state for next session
  • Git checkpoints create recovery points
  • Sub-agents keep heavy work isolated
  • Progressive disclosure preserves context budget

Use what helps. Ignore what doesn't.

This is the knifey-spooney school of project management:

Traditional PMContext Mate
"Follow the methodology""She'll be right"
"Update the Gantt chart"/wrap-session
"Consult the RACI matrix""Oi Claude, what next?"

No ceremonies. No standups with your AI. No burndown charts.

If Homer Simpson can't figure it out in 30 seconds, it's too complicated.

It's all about the context, maaate. 🥄