Daily Report — 2026-05-31
Daily Overview
- What was done: Resolved critical configuration conflicts in the AI coding environment and performed content updates for the personal portfolio website.
- How it was done: By analyzing environment variables, editing JSON configuration files, updating .gitignore rules, and executing deployment scripts via CLI tools.
- Impact: Secured access to advanced LLM capabilities via Claude Platform on AWS and ensured accurate public presentation of professional credentials.
MacBook
- What was done: Investigated Claude Code login failures and initiated the migration from AWS Bedrock to Claude Platform on AWS.
- How it was done: Analyzed CLI output, consulted documentation, modified settings.json to include CLAUDE_CODE_USE_ANTHROPIC_AWS, and verified via smoke tests.
- Impact: Established a stable, modern LLM backend for AI-assisted development, eliminating dependency on legacy Bedrock routing.
TzJsDesktop
- What was done: Updated resume content (PDF/Docx) on the gh-pages site and fixed configuration scripts for cross-platform device sync.
- How it was done: Parsed PDF content, edited Markdown files, adjusted .gitignore patterns to allow resume uploads, and patched setup scripts for universal compatibility.
- Impact: Ensured professional portfolio accuracy and created reliable automation tools for managing AI configurations across multiple machines.
Successfully migrated Claude Code environment from AWS Bedrock to Claude Platform on AWS and updated the personal website resume with latest research.
Tasks
Architecture & Strategy
- ✅ Migrate Claude Code to Platform on AWS — Switched AI provider from AWS Bedrock to Claude Platform on AWS by updating environment variables and settings.json, verified via CLI smoke test.
- ✅ Repair Setup Scripts — Patched PowerShell and Bash setup scripts to support dynamic workspace IDs, API key input, and cross-platform execution for syncing configurations across devices.
Implementation & Fixes
- ✅ Update Portfolio Resume — Integrated latest resume files (PDF/DOCX), updated English and Chinese Markdown versions with new research papers (RecoverBench, Latent Bridge), and deployed to GitHub Pages.
Problems & Solutions
Critical Issues
1. Claude Code refused to switch providers due to conflicting environment variables (CLAUDE_CODE_USE_BEDROCK=1 still active in session history) and missing CLAUDE_CODE_USE_ANTHROPIC_AWS flag.
Solution: Identified the specific variable conflicts via env inspection, removed the Bedrock override, added the Platform on AWS flag to settings.json, and forced a clean restart with verification scripts.
Key Insight: Environment variables in persistent shell profiles can override JSON configuration files; explicit session clearing is required for configuration changes to take effect immediately.
General Issues
2. Git ignored resume PDF/DOCX files during deployment because they fell under broader wildcard patterns in .gitignore (website/static/pdfs/).
Solution: Modified .gitignore to exclude the directory but explicitly whitelist the specific resume files using negation patterns (!filename), then re-staged and committed.
Key Insight: Git’s ignore rules apply recursively; directory-level ignores mask file-level negations unless using wildcard patterns (/*) in the ignore rule itself.
3. Deployment script update.ps1 failed due to incorrect working directory context when invoked from a parent directory, causing path resolution errors for config.yml.
Solution: Wrapped the execution in a PowerShell command that explicitly sets the location to the specific website project root before invoking the script.
Key Insight: Shell scripts relying on relative paths are fragile; explicit context setting is necessary when orchestrating them from different directories.
Human vs AI Approaches
Strategic Level
Configuration Migration Strategy
| Role | Approach |
|---|---|
| Human | The user requested a migration away from Bedrock and asked for environment setup guidance, showing intent to leverage direct Anthropic API features. |
| AI | AI proactively searched documentation, identified the specific env var (CLAUDE_CODE_USE_ANTHROPIC_AWS), and generated cross-platform scripts to automate the change. |
Difference Analysis: Human defined the goal; AI provided the technical path. The AI correctly identified that workspace ID was sufficient without an API key due to existing AWS IAM setup, a nuance not immediately obvious.
Implementation Level
Resume Name Correction
| Role | Approach |
|---|---|
| Human | N/A |
| AI | The AI initially processed the PDF but missed the specific Chinese surname difference, assuming standard translation or OCR output was correct without cross-referencing the user’s direct input preference for accuracy. |
Difference Analysis: Human provided precise semantic correction that AI did not infer from context; demonstrates need for explicit human oversight on culturally sensitive data.
AI Limitations
General Limitations
- AI failed to auto-detect the Chinese surname error in the resume until manually corrected by the user, indicating a gap in verifying culturally specific names against explicit user input.
- Initial attempts to verify AWS configuration resulted in timeouts due to large context windows or hook execution delays, requiring simplified smoke tests.
Learnings
Key Learnings
- When migrating cloud-based LLM providers, always check for conflicting environment variables in the current shell session (Env: variables) as they take precedence over JSON settings files.
- Git’s .gitignore wildcard patterns (e.g., folder/) prevent negation (!file) from working on individual files within that folder; use specific file patterns or directory-specific ignores to fix this.
Practical Learnings
- Automated deployment scripts often fail due to cwd mismatches; always ensure explicit directory changes before running build/utility scripts in an AI-assisted workflow.
Conversation Summaries
AI Environment Setup
✅ Login Failures and Config Migration Initiation 00:27:22.996 | claude_code User attempted to login via Claude Code CLI but encountered authentication errors due to not being logged in. Later, the user requested a migration from AWS Bedrock to Claude Platform on AWS. The AI searched documentation and identified the need for CLAUDE_CODE_USE_ANTHROPIC_AWS=1.
Portfolio Website Management
🔄 Resume Update and Gitignore Fix 23:43:22.654 | claude_code User updated resume files (PDF/DOCX) and requested an update to the website content on device TzJsDesktop. The AI parsed the PDF, updated English and Chinese Markdown files, and attempted to commit. It encountered a Git ignore conflict for the /pdfs directory.
Portfolio Website Deployment
✅ Git Fix and Deployment Completion 23:43:22.654 | claude_code AI corrected the .gitignore to allow resume files, committed changes, and pushed to GitHub. The user corrected the Chinese name from 唐 to 汤. A deployment script was run with path fixes to successfully publish the site.
AWS IAM and Config Verification
✅ Detailed AWS Config Audit 00:07:38.619 | codex On TzJsDesktop, the user asked for a status check on reaching Claude Platform on AWS. Codex performed a deep audit of environment variables, IAM policies (AdministratorAccess), and existing settings. It identified that CLAUDE_CODE_USE_BEDROCK was still active in the shell history.
Script Repair and Verification
✅ Setup Script Patching 00:07:38.619 | codex AI edited setup-claude-platform-aws.ps1 and .sh to add parameters for WorkspaceId and API Key, removing hardcoded values. It verified syntax and executed the PowerShell script successfully, confirming AWS credentials validity.