Environment Detection

"Know your battlefield before deploying your troops."

TL;DR

Detects the runtime environment (IDE + model) and adapts Pristino's behavior accordingly. Determines triad mode (full/sequential/checklist/suggestion), skill loading strategy, and context budget. Runs at session start as part of the bootstrap sequence. Full protocol: references/ontology/environment-protocol.md. [EXPLICIT]

Procedure

Step 1: Discover

• Check which instruction file loaded the agent (CLAUDE.md, GEMINI.md, .cursorrules, etc.)
• Check available tools (Agent tool present? → full triad mode)
• Check model context window (how much can we load?)
• Read references/ontology/environment-protocol.md for the full detection matrix

Step 2: Analyze

• Map IDE to capability profile:
  • claude-code → full triad (subagents, hooks, MCP)
  • gemini/antigravity → sequential prompts (no Agent tool)
  • cursor/windsurf → checklist mode (inline rules)
  • copilot → suggestion mode (limited context)
  • codex → sequential prompts (Bash/Read/Write only)
• Map model to tier:
  • Heavy (>100K): load full Constitution + Index + history
  • Medium (32-100K): load Constitution + active skills
  • Light (<32K): load active skill only

Step 3: Execute

• Set triad mode based on IDE detection
• Set context budget based on model tier
• Load appropriate files for the tier
• Report environment to user:

  Environment detected:
    IDE: {ide} | Model: {model} | Tier: {tier}
    Triad: {mode} | Tools: {available_tools}
    Skills: {count} | Agents: {count}
  

Step 4: Validate

• Confirm detected IDE matches actual tool availability
• Confirm model tier matches actual context capacity
• If mismatch: adjust and re-report
• Log detection result for session continuity

Quality Criteria

• IDE correctly identified from instruction file
• Model tier correctly assessed
• Triad mode matches IDE capabilities
• Context budget set appropriately
• Environment reported to user
• Evidence tags applied

Anti-Patterns

Anti-Pattern	Why It's Bad	Do This Instead
Assuming Claude Code always	May be running in Cursor/Copilot	Detect from instruction file
Loading full Index in Light tier	Exhausts context window	Load only active skill
Skipping detection	Triad mode wrong for the IDE	Always detect at session start
Hardcoding model name	Models change, tiers don't	Detect by capability, not name

Related Skills

• session-protocol — Environment detection is Step 0 of session init
• context-optimization — Context budget depends on detected tier
• continuous-learning — Log environment patterns for optimization

Usage

Example invocations:

• "/environment-detection" — Run the full environment detection workflow
• "environment detection on this project" — Apply to current context

Assumptions & Limits

• Assumes access to project artifacts (code, docs, configs) [EXPLICIT]
• Requires English-language output unless otherwise specified [EXPLICIT]
• Does not replace domain expert judgment for final decisions [EXPLICIT]

Edge Cases

Scenario	Handling
Empty or minimal input	Request clarification before proceeding
Conflicting requirements	Flag conflicts explicitly, propose resolution
Out-of-scope request	Redirect to appropriate skill or escalate