5. Who owns the evidence, and what fails quietly? — Agent Workspace Practice MCP Resource

• Claude Code Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? run in your project directory
• Claude Desktop Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? ~/Library/Application Support/Claude/claude_desktop_config.json
• Cursor Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? ~/.cursor/mcp.json
• VS Code Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? .vscode/mcp.json
• Zed Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? ~/.config/zed/settings.json
• Windsurf Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? ~/.codeium/windsurf/mcp_config.json
• Cline Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? ~/Library/Application Support/Code/User/globalStorage/saoudrizwan.claude-dev/settings/cline_mcp_settings.json
• Gemini CLI Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? ~/.gemini/settings.json
• Grok Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? .mcp.json (in your project root)
• ChatGPT Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? Settings → Connectors → Advanced → Developer mode
• Claude.ai Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? Settings → Connectors → Add custom connector
• LangChain Agent Workspace Practice Resource/5. Who owns the evidence, and what fails quietly? pip install langchain-mcp-adapters

Other resources on this server

• Patterns
• 1. Pure roles, composed with project facts
• 2. Classify-then-act, not ask-then-wait
• 3. Make silent failure loud (the dead-man's switch)
• 4. Tier by mechanical impact, not by tone
• 5. Memory points, it doesn't mirror
• 6. Credentials live in one place, never in files
• 7. A cheap hook beats a careful agent
• 8. Audit the workspace like a fitness function
• 9. Context is a budget, not a constant
• 10. A skill is editable weights — never adopt a self-edit without a gate
• 11. A scaffold is a hypothesis — gate it behind a measurable signal
• 12. Loop selection: not everything should be a loop
• 13. Challenge half-formed ideas with a different lens — and hold a samp…
• 14. Delegation is a queue you fill, not work the agent finds
• 15. Price the lane before you migrate it
• 16. A claim carries its provenance, or it is a guess
• 17. One canonical copy, and pointers from everywhere else
• 18. Position is price — a token costs more the earlier you add it
• How they compose
• Evaluation: knowing whether a workspace change helped
• The instrument
• 1. Deterministic checks only, never a judge
• 2. A pass-rate with variance, never a single number
• 3. Five fixture classes, and two of them are the point
• What it costs
• How it connects to the rest
• Where it lives
• M0. Foundations — the session loop
• Do this
• Measure it
• M1. Canonical knowledge — one source of truth
• The patterns
• M2. Context economics — context is spend
• M3. Verification & oversight — trust through checks
• M4. Safety & permissions — cheap mechanical guards
• M5. Telemetry & cost — loud failure, priced lanes
• M6. Provenance & delegation — sourced claims, mandated work
• Learn the architecture
• Where to start