Critical behavioral feedback from 26-session enterprise project (sent via desktop app)

We submitted detailed feedback through the Claude Desktop application on 2026-02-26 containing findings from a 26-session enterprise software development project.

The feedback includes 12 critical behavioral findings and 12 strengths, all backed by real production evidence from a multi-agent environment (Claude Opus + OpenAI Codex + custom agent).

Why this matters: These behavioral patterns have safety implications beyond software quality — particularly for critical infrastructure modernization scenarios.

We chose not to publish the details publicly as the findings could be exploited for prompt injection and social engineering.

Request: Please route the desktop app feedback from this date to the appropriate safety/alignment team for review.

Account context: Claude Code + Claude Desktop, enterprise .NET project, Windows environment.