DeepExtractRuntime: Instructions file for Claude Code

DeepExtractIDA Agent Analysis Runtime

AI-driven binary analysis runtime that turns IDA Pro decompiled code and SQLite analysis databases into structured, queryable intelligence. Analyzes Windows PE binaries via slash commands, specialized agents, analysis skills, and shared helper modules.

Quick Rules

• All extraction databases are read-only. Never write to them.
• Always use helpers for DB access, function resolution, error handling, and classification -- never reimplement what helpers/ provides.
• Read a skill's SKILL.md before invoking it for the first time.
• Use --json when capturing script output programmatically; data to stdout, progress/errors to stderr.
• After resolving a function, use --id <function_id> in all subsequent calls.
• Prefer cached results from cache/. Never pass --no-cache unless the user explicitly asks.
• Degrade gracefully on missing data -- never crash with opaque errors; explain what is missing.
• Never suppress stderr when running skill scripts. Do not use 2>/dev/null or equivalent redirections -- structured errors and diagnostics are emitted on stderr by design. Suppressing them hides the exact information needed to debug failures.
• Before creating or modifying a command, skill, or agent, read the corresponding authoring guide in docs/ and follow its checklist.

Workflow Principles

• Plan before building. Enter plan mode for any non-trivial task (3+ steps or architectural decisions). Write detailed specs upfront to reduce ambiguity. If something goes sideways, stop and re-plan immediately.
• Use subagents liberally. Offload research, exploration, and parallel analysis to subagents. One task per subagent for focused execution. For complex problems, throw more compute at it via subagents.
• Verify before marking done. Never mark a task complete without proving it works. Diff behavior between main and your changes. Run tests, check logs, demonstrate correctness. Ask yourself: "Would a staff engineer approve this?"
• Demand elegance. For non-trivial changes, pause and ask "is there a more elegant way?" If a fix feels hacky, implement the clean solution. Challenge your own work before presenting it.
• Fix bugs autonomously. When given a bug report, just fix it. Point at logs, errors, failing tests -- then resolve them. Zero context switching required from the user.
• Simplicity first, minimal impact. Make every change as simple as possible. Find root causes, not temporary fixes. Changes should only touch what is necessary.