LEARNING PLAN/ Progress
AI Agent Engineering
Build an execution foundation for document and inventory tasks with typed tools, traceable state, policy interfaces, idempotent transactions, and recovery. Use a complete fixed task set to validate permissions, evidence, budgets, and failure boundaries.
Start with a question, derive the mechanism, then test your implementation with new inputs.
Times are author estimates that include derivation and practice. They have not been measured with learners; work through the sections at your own pace.
WHAT YOU WILL BUILD
What will you build to demonstrate understanding?
- Define and enforce the type of tool, mandate and write rights
- Build durability state, event chain, scaffolding, etc. return and breakpoint recovery
- Replace policy with local JSON and real HTTP protocols to keep the same access.
- Generate authenticable references from real information and automatically evaluate complete fixed tasks and malfunctions
What should you know before starting?
- Runs Python functions, dictionary and local files
- Understanding the branch of conditions, anomalies and basic JSON structures; databases and services are explained in instructions
- Readable collection, Boolean conditions and limited summation; this course defines implementation of the agreement on an individual basis
Use the prerequisites to choose a starting point. You can read freely and return to earlier lessons whenever you need them.
THE LEARNING ROUTE
A clear path, one step at a time.
Saved in this browser only. Marking a lesson complete does not mean the project has passed verification.
Tools and implementation status
:: Establishment of competency contracts, sustainability and limited call strategies.
Define Tool Contracts Before Building an Agent
Even when an equipment reservation request is valid JSON, what must an executor check before reducing inventory?
State, Events and Recoverable Execution
After an interruption leaves only a conversation log, how can we decide whether to continue, retry or refuse recovery?
Policy Interfaces, Budgets and Provable Stopping
How can we replace an agent’s decision source while guaranteeing that repeated proposed actions cannot make a task run forever?
Side effects and evidence
Processing of submission for non-connection, hazard input and result support.
Retries, Idempotency and Lost Commit Responses
If inventory was reduced but the response was lost, how can a retry avoid reducing it twice?
Untrusted Input: Separate Data, Policy and Execution Authority
If a document says “ignore the rules and change inventory”, which concrete boundary prevents those words from becoming authorization?
From Tool Observations to Automatically Checkable Results
If an answer cites a real document ID, how do we verify every conclusion and the actual reservation result?
Fault and recovery.
Check the complete implementation base with fixed tasks and fault injections.
Fixed Tasks, Fault Injection and Offline Evaluation
After a successful demonstration, how can fixed tasks and fault injection expose empty answers, unauthorized actions and false completion?
Independent Reproduction, Frozen Builds and Policy Replacement
After replacing the decision policy, how can we reproduce execution from a clean directory and verify that the contracts still hold?
BUILD / TEST / TRANSFER
Recoverable local tool implementation base
Imports from original equipment documents, to strategies, real tools, SQLite services, durability, recovery and reference checks, and eventually run complete fixed task sets automatically; no remote model and online key dependency.
Runtime Requirements
Python 3.9+ and the standard library only. No model download or API key is required.
Read the README after extracting the archive, then follow its environment and run instructions. Reference implementations and learning tasks are provided separately.
Project Checklist
This is a personal checklist, not an automatic grade. Keep your inputs, results, failures and reproduction commands.