Agent Workspace Practice MCP Resources

Agent Workspace Practice offers 71 resources. A resource is data the server exposes for a client to attach as context — a file, a page, an API response — rather than an action. Each has its own page, per client.

• 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
• The modules
• How each module works
• Track your progress
• Questions and corrections
• 1. Where does the organisation live, as far as the agent can see?
• 2. What does the organisation keep from AI work, and what does each ses…
• 3. How does AI output earn its way into a decision?
• 4. What can it not do, and how do you know?
• 5. Who owns the evidence, and what fails quietly?
• 6. Which path is easier, the governed one or the shadow one?
• Six decisions before your agents scale
• How to use it
• The briefs
• Where the evidence comes from
• The 30-minute version
• Teardown: 12-Factor Agents
• What it is
• What works
• The trade-offs
• What's conspicuously absent
• What this teaches
• Teardown: herdr
• Teardown: LifeOS
• Teardown: DeepSeek Harness
• Teardowns
• Why teardowns
• Ground rules
• Page conventions
• Distribution
• Published
• Disagree with a reading?

Read from the server itself, by connecting to it and calling resources/list on 24 September 2026. A listing does not declare its resources, so asking is the only way to know them, and this is what the server actually offers rather than what its listing claims. Names only are stored; connect the server for each resource's type and contents.