202021035
Tool: Codex CLI (Windows)
Resolved PowerShell execution policy error (Set-ExecutionPolicy), Windows environment variable setup (export → set). Systematic troubleshooting report.
Assignment Submissions
This page summarizes student assignments submitted via GitHub PRs. See the Contributing Guide for submission instructions.
Submission Overview
Section titled “Submission Overview”Weekly Assignments (Week 01–08)
Section titled “Weekly Assignments (Week 01–08)”| Student ID | W01 | W02 | W03 | W04 | W05 | W06 | W07 | W08 |
|---|---|---|---|---|---|---|---|---|
| 202021016 | Submitted | |||||||
| 202021035 | Submitted | Submitted | ||||||
| 202121014 | Submitted | Submitted | ||||||
| 202321005 | Submitted | Submitted | Submitted | |||||
| 202321006 | Submitted | Submitted | Submitted | Submitted | Submitted | Submitted | Submitted | Submitted |
| 202321010 | Submitted | Submitted | Submitted | Submitted | Submitted | Submitted |
Labs and Capstone
Section titled “Labs and Capstone”| Student ID | Lab 02 | Lab 03 | Lab 05 | Lab 06 | Lab 07 | Capstone |
|---|---|---|---|---|---|---|
| 202121014 | Submitted | Submitted | Submitted | Submitted | Submitted | |
| 202321005 | Submitted | |||||
| 202321006 | Submitted | |||||
| 202321010 | Submitted |
- Submitted Meets assignment requirements
- Submitted Submitted but needs supplementation
Week 01: AI Coding CLI Installation and Execution
Section titled “Week 01: AI Coding CLI Installation and Execution”Assignment: Install an AI coding CLI (Claude Code, Gemini CLI, or Codex CLI), create hello_agent.py, document installation issues and solutions
202121014
Tool: Codex CLI
Verified Node.js version compatibility (v24 vs recommended v20). Selected ChatGPT account login for authentication.
202321005
Tool: Claude Code
Resolved SSH Permission denied (ssh-keygen + ssh-copy-id). Fixed CLI PATH not registered issue.
202321006
Tool: Codex CLI
Implemented a Pygame pinball game. Resolved Python 3.14 distutils removal issue using uv run --python 3.12.
202321010
Tool: Gemini CLI
Batch submission of Weeks 01–04. Includes installation screenshots and hello_agent.py.
Week 02: Governance and MIG Simulation
Section titled “Week 02: Governance and MIG Simulation”Assignment: HOTL governance, MIG virtual simulation
202321006
Designed virtual MIG data structures with FastMCP. Overcame MIG non-support on RTX 3060 through simulation. Implemented TBAC-based role access control (Professor/Student).
Week 03: MCP Server Implementation and Security Verification
Section titled “Week 03: MCP Server Implementation and Security Verification”Assignment: FastMCP server implementation (Tool + Resource + Prompt), MIG profile analysis, governance gateway
202321005
MCP server (mcp_server.py) + Governed Gateway (mcp_gateway.py). Includes MIG analysis report, K8s nodeAffinity example, Llama-3-8B benchmark, TBAC/SANDWORM security reports, MCP Inspector capture JSONs. Most comprehensive submission.
202321006
MIG profile partition analysis (Strategy A vs B comparison), TBAC 3-layer architecture (Mermaid diagrams), McpInject attack simulation and defense strategies, Llama-3-8B 4-bit VRAM calculation (~6GB), mig_monitor_server.py JSON-RPC verification.
202321010
MCP GPU server, MIG profile analysis, architecture documentation, K8s Pod configuration.
Week 04: Ralph Loop Implementation and Test-time Compute Scaling
Section titled “Week 04: Ralph Loop Implementation and Test-time Compute Scaling”Assignment: Ralph loop harness (harness.sh) implementation — backpressure, garbage collection, AGENTS.md cumulative learning
202021016
Insufficient content. README is a copy of the project root README, AGENTS.md contains only “CLAUDE.md”. No harness.sh or loop implementation submitted. Needs supplementation.
202021035
harness.sh (159 lines) + run_task.py. tasks.json-based task queue, 3-iteration progress archive, metrics.csv collection. Clean separation of AGENTS.md and PROMPT.md for role clarity.
202121014
harness.sh (103 lines), backpressure.py, AGENTS.md (Learned Patterns + Anti-Patterns + Progress checklist), PROMPT.md, 805-line execution log. Records divide-by-zero handling learning process for calculator task.
202321005
Most sophisticated implementation. harness.sh (335 lines), backpressure.py (149 lines), mock_agent.py, 4 task checkpoints, 7 pytest error logs, loop metrics (CSV+JSON), worktree analysis report, RLM chunk demo experiment.
202321006
harness.sh (76 lines) — backpressure (pytest exit code), garbage collection (file deletion), stuck detection (task splitting after 2 consecutive failures), loop metrics JSON recording. All 5 loops failed due to pytest: command not found — environment setup issue.
202321010
autoresearch.py (167 lines), harness.sh (437 lines) — longest harness script. Applied JSON feature list + Initializer pattern. Includes loop results JSON, metrics CSV, worktree report.
Week 05: Context Management and Token Optimization
Section titled “Week 05: Context Management and Token Optimization”Assignment: Context Manager, token counter, Context Rot measurement and auto-reset
202321006
context_manager.py dynamic context pruning + priority-based filtering. token_counter.py accurate token counting for Claude models. Context Rot mitigation via sliding window.
202321010
Token counter-integrated Ralph loop, Context Rot simulation and measurement, Hybrid auto-context reset. fix_plan.md + claude-progress.txt state tracking. Analysis scripts with before/after visualization graphs (5 types).
Week 06: Instruction Tuning and Log Analysis
Section titled “Week 06: Instruction Tuning and Log Analysis”Assignment: Prompt improvement experiments, log analysis, Instruction Tuning impact measurement
202321006
Added architecture constraints to PROMPT.md, achieving 30% reduction in logic errors. log_analyzer.py tracks agent reasoning paths, identifies common failure points. Confirmed stabilization effect of explicit CLAUDE.md rules.
202321010
Instruction Tuning experiment — testset + before/after output comparison. Generated error rate graph, metrics graph, error type classification graph. Systematic organization under lab06 subdirectory.
Week 07: Multi-Agent SDLC Design
Section titled “Week 07: Multi-Agent SDLC Design”Assignment: 5-stage pipeline multi-agent architecture, JSON artifact-based communication, verification gates
202321006
5-phase gated pipeline architecture diagram, DAG-based task decomposition + tiered parallelization strategy, phase transition quality checklists, recovery strategies for 3 failure scenarios. 4 schemas (requirement.json, task.json, pipeline_state.json, lesson.json).
202321010
Planner → Context → Builder → Reviewer → Finalizer 5-stage pipeline. 4+ standard JSON schemas, DAG dependency graph + parallelization tier analysis, 5 gate checklists per phase, automatic recovery mechanisms for 3 scenarios.
Week 08: Capstone Proposals
Section titled “Week 08: Capstone Proposals”Assignment: Capstone project proposal — problem definition, system design, milestones
202321005 — Socratic Tutor
Educational agent that corrects student misconceptions through questions alone, never revealing answer code. Dual-loop · 4-module · 3-tier defense architecture with Analysis/Dialogue/Review/Logging modules, Q-Critic + Validator parallel AND gate.
202321006 — Collaboration Nudge Agent
Real-time bus factor calculation on GitHub PR events to detect knowledge monopoly. Extracts design context via AI interview, redistributes knowledge to suitable team members. Prevents problems typically discovered only after employee departure.
202321010 — Docs-Code Drift Detector
AST-based comparison of type/parameter definitions in READMEs, docstrings, and API docs against actual code to detect drift. Auto-generates documentation fix PRs (code fixes are suggestions only).
Lab 02: HOTL Governance Agent
Section titled “Lab 02: HOTL Governance Agent”Assignment: Combine HOTL governance layer + audit logging on an Anthropic SDK-based agent
202121014
governance.py 4-tier risk classification + approval policy, agent.py Claude suggestion/fallback passthrough, pytest tests, JSONL audit log sample. Documented design decisions.
Lab 03: FastMCP Server Implementation
Section titled “Lab 03: FastMCP Server Implementation”Assignment: FastMCP-based Tool + Resource + Prompt primitives, Path Traversal defense, safe subprocess policy
202121014
filesystem/git/custom server registration (settings.json), custom MCP server (custom_server.py), 3 pytest samples. Covers baseline security requirements.
Lab 05: Context Management Lab
Section titled “Lab 05: Context Management Lab”Assignment: Token tracking, Rolling Window context compression, state save/restore
202121014
python main.py runs full demo automatically. Token usage tracking, Rolling Window compression, execution state save/restore, auto-generation of claude-progress.txt.
Lab 06: Prompt Improvement and A/B Test
Section titled “Lab 06: Prompt Improvement and A/B Test”Assignment: Analyze harness.log, extract recurring failure patterns, A/B compare two prompt versions
202121014
log_analyzer.py classifies recurring log error patterns + generates error_report.md, compares prompt v1 (baseline) and v2 (improved), ab_test.py records results to ab_results.json.
Lab 07: Multi-Agent Pipeline
Section titled “Lab 07: Multi-Agent Pipeline”Assignment: Planner → Coder 2-stage pipeline with JSON schema-based structured outputs
202121014
Planner/Coder output JSON schemas, base_agent.py common class, planner_agent.py (goal → subtask decomposition), coder_agent.py (plan → change set), pipeline.py sequential connection. Design documentation included.