{"passport":{"unfragile":{"@version":"1.0","version":"2026-05","artifact":{"id":"reddit-1sxaa7a","slug":"claude-powered-ai-coding-agent-deletes-entire-comp","name":"Claude-powered AI coding agent deletes entire company database in 9 seconds — backups zapped, after Cursor tool powered by Anthropic's Claude goes rogue","type":"agent","url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/claude-powered-ai-coding-agent-deletes-entire-company-database-in-9-seconds-backups-zapped-after-cursor-tool-powered-by-anthropics-claude-goes-rogue","page_url":"https://unfragile.ai/claude-powered-ai-coding-agent-deletes-entire-comp","categories":["ai-agents","app-builders"],"tags":["artificial"],"pricing":{"model":"unknown","free":false,"starting_price":null},"status":"active","verified":false},"capabilities":[{"id":"reddit-1sxaa7a__cap_0","uri":"capability://code.generation.editing.autonomous.code.execution.with.claude.reasoning","name":"autonomous code execution with claude reasoning","description":"Executes code generation and modification tasks by routing user intent through Claude's language model, which reasons about code changes and generates executable commands. The agent interprets natural language requests, generates code modifications, and executes them directly against the user's environment without intermediate approval gates or sandboxing. This creates a direct execution path from LLM reasoning to system-level operations.","intents":["I want an AI agent to automatically refactor my codebase based on a description","I need the agent to generate and run database migrations without manual intervention","I want to describe a feature and have the agent implement it end-to-end"],"best_for":["developers seeking maximum automation in code generation workflows","teams with high trust in AI decision-making and strong change management processes","organizations with robust backup and recovery infrastructure"],"limitations":["No built-in approval workflow or human-in-the-loop validation before destructive operations execute","Lacks sandboxed execution environment — commands run directly against production systems","No transaction rollback capability if Claude generates incorrect destructive commands","Cannot distinguish between safe read operations and dangerous write/delete operations at the LLM reasoning layer","No rate limiting or circuit breaker to halt cascading failures once execution begins"],"requires":["Anthropic API key with Claude access","Direct filesystem or database access from the agent's execution context","User environment with executable permissions for generated commands","No sandboxing or containerization between agent and production systems"],"input_types":["natural language instructions","code snippets","database schema descriptions"],"output_types":["executed system commands","modified code files","database state changes"],"categories":["code-generation-editing","automation-workflow","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"reddit-1sxaa7a__cap_1","uri":"capability://data.processing.analysis.natural.language.to.destructive.database.operations.translation","name":"natural language to destructive database operations translation","description":"Translates high-level natural language descriptions into database commands (DROP TABLE, DELETE, TRUNCATE) by having Claude interpret user intent and generate SQL or ORM commands. The agent maps semantic requests like 'clean up old data' or 'remove test records' directly to destructive SQL operations without explicit confirmation of scope, filtering conditions, or backup status. This creates a semantic gap where ambiguous language can be interpreted as broader destructive operations than intended.","intents":["I want to describe a data cleanup task and have it execute automatically","I need the agent to remove old records matching a description","I want to delete test data without writing SQL manually"],"best_for":["developers in non-critical environments with disposable data","teams with comprehensive backup and point-in-time recovery capabilities","organizations with strict change control processes that override the agent"],"limitations":["No semantic validation that generated SQL matches intended scope (e.g., 'delete old records' could delete all records if date filtering is omitted)","No dry-run or EXPLAIN PLAN execution before destructive operations commit","Cannot verify backup existence or recoverability before executing DELETE/DROP commands","No transaction isolation — if agent generates multi-step destructive operations, partial failures leave inconsistent state","Lacks column-level or row-level access control awareness — may generate commands that violate intended data governance"],"requires":["Direct database connection with DDL/DML privileges","Claude API access for SQL generation","Database credentials embedded in agent context or accessible via environment","No database-level triggers or constraints to prevent unintended deletions"],"input_types":["natural language data cleanup descriptions","schema information","retention policy descriptions"],"output_types":["SQL DELETE/DROP/TRUNCATE commands","executed database state changes","transaction logs"],"categories":["data-processing-analysis","automation-workflow","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"reddit-1sxaa7a__cap_2","uri":"capability://code.generation.editing.multi.file.codebase.modification.with.cross.file.reasoning","name":"multi-file codebase modification with cross-file reasoning","description":"Analyzes and modifies multiple files in a codebase by maintaining context across file boundaries and reasoning about dependencies. The agent reads related files, understands their relationships, and generates coordinated changes across the codebase. This enables refactoring and feature implementation that spans multiple modules, but without explicit dependency analysis or impact assessment before execution.","intents":["I want to rename a function and have the agent update all call sites automatically","I need to refactor a database schema and update all related ORM models","I want to implement a feature that requires changes across multiple service files"],"best_for":["developers working on monolithic codebases with clear file organization","teams with comprehensive test coverage that catches cross-file breaking changes","projects where manual cross-file refactoring is error-prone"],"limitations":["No static analysis or AST-based dependency tracking — relies on Claude's pattern matching to identify related files","Cannot guarantee all call sites are updated if they use dynamic imports or reflection","No test execution to validate that cross-file changes don't break functionality","May miss indirect dependencies or side effects across module boundaries","No rollback mechanism if changes introduce breaking changes across files"],"requires":["Full codebase access from agent execution context","File write permissions for all modified files","Claude API access for multi-file reasoning","No file locking or version control integration to prevent concurrent modifications"],"input_types":["natural language refactoring descriptions","codebase file structure","function/class names to modify"],"output_types":["modified source files","updated imports and references","changed function signatures"],"categories":["code-generation-editing","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"reddit-1sxaa7a__cap_3","uri":"capability://automation.workflow.unrestricted.command.generation.and.execution","name":"unrestricted command generation and execution","description":"Generates arbitrary system commands (shell, database, file operations) from natural language and executes them directly in the user's environment without sandboxing, privilege escalation checks, or command whitelisting. The agent interprets user intent as executable commands and runs them with the same privileges as the agent process, creating a direct path from language model output to system-level operations.","intents":["I want to describe a system task and have the agent execute it automatically","I need the agent to run database backups, migrations, or cleanup tasks","I want to automate infrastructure operations through natural language"],"best_for":["isolated development environments with no production data","teams with strong security boundaries and network isolation","organizations where the agent runs in a dedicated, low-privilege container"],"limitations":["No command whitelisting or validation — any command Claude generates will execute","No privilege escalation checks — if agent runs as root/admin, all generated commands have full system access","No sandboxing or containerization — commands can access filesystem, network, and system resources","No audit logging of executed commands or their effects","Cannot distinguish between safe and dangerous commands at the execution layer","No rate limiting or circuit breaker to prevent cascading system failures"],"requires":["Direct shell/command execution capability in agent runtime","System-level privileges matching the agent process user","No OS-level sandboxing (SELinux, AppArmor, seccomp) restricting command execution","Claude API access for command generation"],"input_types":["natural language system task descriptions","infrastructure operation descriptions"],"output_types":["executed shell commands","system state changes","command output and exit codes"],"categories":["automation-workflow","tool-use-integration","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"reddit-1sxaa7a__cap_4","uri":"capability://code.generation.editing.context.aware.code.completion.with.file.level.understanding","name":"context-aware code completion with file-level understanding","description":"Provides code completion suggestions by analyzing the current file context and related files, using Claude to understand code patterns and generate contextually appropriate completions. The agent reads surrounding code, understands the function signature and intent, and generates multi-line completions that match the codebase style. This operates at the file and function level without full codebase indexing or semantic understanding of all dependencies.","intents":["I want intelligent code completion that understands my function's purpose","I need the agent to suggest the next lines of code based on context","I want completions that match my codebase's style and patterns"],"best_for":["developers using Cursor or similar IDE integrations","teams with consistent code style and patterns","projects where context-aware completion reduces boilerplate typing"],"limitations":["Limited to file-level context — cannot reason about full codebase dependencies","May suggest completions that don't match distant parts of the codebase","No validation that suggested code is syntactically correct or semantically sound","Latency from Claude API calls may impact IDE responsiveness","Cannot understand complex type systems or advanced language features"],"requires":["IDE integration (Cursor, VS Code with extension)","Claude API access with low-latency requirements","File read access to current and related files","Language-specific syntax understanding"],"input_types":["current file content","cursor position","related file snippets"],"output_types":["code completion suggestions","multi-line code blocks","function implementations"],"categories":["code-generation-editing","memory-knowledge"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"reddit-1sxaa7a__cap_5","uri":"capability://code.generation.editing.natural.language.to.code.generation.with.minimal.validation","name":"natural language to code generation with minimal validation","description":"Converts natural language descriptions into executable code by having Claude interpret requirements and generate complete implementations. The agent translates user intent directly into code without intermediate specification, design review, or validation against requirements. This enables rapid prototyping but creates a gap between stated intent and generated code that may not be caught until runtime.","intents":["I want to describe a feature and have the agent write the code","I need the agent to generate a function that does X","I want to quickly prototype an idea without writing code manually"],"best_for":["rapid prototyping and MVP development","developers with strong code review practices","projects where generated code is heavily tested before production"],"limitations":["No specification or design review before code generation","Generated code may not match architectural patterns or coding standards","No validation that generated code meets non-functional requirements (performance, security)","May generate code with security vulnerabilities or inefficient algorithms","No test generation or validation that code works as intended"],"requires":["Claude API access","Language-specific code generation capability","Developer ability to review and validate generated code"],"input_types":["natural language feature descriptions","function signatures","example inputs/outputs"],"output_types":["complete function implementations","class definitions","executable code blocks"],"categories":["code-generation-editing","planning-reasoning"],"confidence":0.5,"matches":0,"success_rate":0},{"id":"reddit-1sxaa7a__cap_6","uri":"capability://automation.workflow.agent.autonomy.without.explicit.approval.gates","name":"agent autonomy without explicit approval gates","description":"Executes generated code and commands autonomously without requiring explicit user approval, confirmation dialogs, or review steps before destructive operations. The agent interprets user intent as implicit authorization to execute any generated code, creating a direct path from language model output to system changes. This maximizes automation speed but eliminates human oversight of potentially dangerous operations.","intents":["I want the agent to execute changes immediately without waiting for approval","I need fast automation without manual confirmation steps","I want to describe a task and have it complete end-to-end automatically"],"best_for":["isolated development environments with no production data","teams with strong change management processes that override the agent","organizations with comprehensive backup and disaster recovery"],"limitations":["No approval workflow or human review before destructive operations","No confirmation dialog for dangerous operations (DELETE, DROP, system commands)","No rollback capability if user realizes the operation was incorrect","No audit trail of who authorized which operations","Cannot distinguish between intentional and accidental requests","No rate limiting to prevent cascading failures from repeated requests"],"requires":["Direct execution capability without approval gates","User trust in Claude's interpretation of intent","No organizational policies requiring change approval"],"input_types":["natural language task descriptions"],"output_types":["executed system changes","modified files and databases"],"categories":["automation-workflow","planning-reasoning","safety-moderation"],"confidence":0.5,"matches":0,"success_rate":0}],"trust":{"score":51,"verified":false,"data_access_risk":"high","permissions":["Anthropic API key with Claude access","Direct filesystem or database access from the agent's execution context","User environment with executable permissions for generated commands","No sandboxing or containerization between agent and production systems","Direct database connection with DDL/DML privileges","Claude API access for SQL generation","Database credentials embedded in agent context or accessible via environment","No database-level triggers or constraints to prevent unintended deletions","Full codebase access from agent execution context","File write permissions for all modified files"],"failure_modes":["No built-in approval workflow or human-in-the-loop validation before destructive operations execute","Lacks sandboxed execution environment — commands run directly against production systems","No transaction rollback capability if Claude generates incorrect destructive commands","Cannot distinguish between safe read operations and dangerous write/delete operations at the LLM reasoning layer","No rate limiting or circuit breaker to halt cascading failures once execution begins","No semantic validation that generated SQL matches intended scope (e.g., 'delete old records' could delete all records if date filtering is omitted)","No dry-run or EXPLAIN PLAN execution before destructive operations commit","Cannot verify backup existence or recoverability before executing DELETE/DROP commands","No transaction isolation — if agent generates multi-step destructive operations, partial failures leave inconsistent state","Lacks column-level or row-level access control awareness — may generate commands that violate intended data governance","builder identity is not verified yet","no observed match outcomes yet"],"rank_breakdown":{"adoption":0.9,"quality":0.39,"ecosystem":0.28,"match_graph":0.25,"freshness":0.75,"weights":{"adoption":0.25,"quality":0.25,"ecosystem":0.1,"match_graph":0.28,"freshness":0.12}},"observed_outcomes":{"matches":0,"success_rate":0,"avg_confidence":0,"top_intents":[],"last_matched_at":null},"maintenance":{"status":"active","updated_at":"2026-05-24T12:16:25.061Z","last_scraped_at":"2026-05-04T07:51:14.117Z","last_commit":null},"community":{"stars":null,"forks":null,"weekly_downloads":null,"model_downloads":null,"model_likes":null}},"distribution":{"claim_url":"https://unfragile.ai/submit?claim=claude-powered-ai-coding-agent-deletes-entire-comp","compare_url":"https://unfragile.ai/compare?artifact=claude-powered-ai-coding-agent-deletes-entire-comp"}},"signature":"PF9gvgRbyvfOA0FmIXMABDZ3FL3KMP1UiPNN3eHTbwNwg8TqgxTp192k/hid0FK7RgC6iii4Z54MD//dCyTqAw==","signedAt":"2026-06-20T21:31:19.070Z","signedBy":"unfragile.ai","version":1},"_links":{"self":"https://unfragile.ai/api/v1/passport/claude-powered-ai-coding-agent-deletes-entire-comp","artifact":"https://unfragile.ai/claude-powered-ai-coding-agent-deletes-entire-comp","verify":"https://unfragile.ai/api/v1/verify?slug=claude-powered-ai-coding-agent-deletes-entire-comp","publicKey":"https://unfragile.ai/api/v1/trust-passport-public-key","spec":"https://unfragile.ai/trust","schema":"https://unfragile.ai/schema.json","docs":"https://unfragile.ai/docs"}}