reverse-engineering-deep-analysis

Purpose

Perform extended, high-effort reverse engineering that blends static/dynamic/symbolic analysis and deobfuscation to reach hard paths and validate exploitability. Built on skill-forge structure-first rules and prompt-architect explicit constraints/confidence ceilings.

Use When / Redirect When

• Use when: complex obfuscation, layered packers, VM-based protections, or exploit research requiring end-to-end proof.
• Redirect when: firmware focus (reverse-engineering-firmware), quick IOC triage (reverse-engineering-quick), or lighter triage (reverse-engineer-debug).

Guardrails

• Isolated sandboxes with snapshots; controlled network (block or explicit allowlist).
• Preserve chain-of-custody (hashes, timestamps, env/tool versions).
• Avoid external service uploads without approval.
• Confidence ceilings enforced (inference/report ≤0.70, research 0.85, observation/definition 0.95).

Prompt Architecture Overlay

1. HARD/SOFT/INFERRED constraints (arch, protections, goals, time/effort budget, tooling).
2. Two-pass refinement: structure → epistemic.
3. English-only output with explicit confidence line.

SOP (Extended Deep Loop)

1. Scope & Toolchain: Confirm authorization, hashes, targets, and success criteria; select deobfuscation, unpacking, tracing, and symbolic tools.
2. Preparation: Detect packers/anti-debug/VM tricks; set breakpoints/hooks; craft harnesses and inputs.
3. Execution: Iteratively unpack/decrypt stages, trace control/data flows, and capture memory dumps.
4. Exploit Research: Validate reachability, build PoCs, and map mitigations (ASLR/DEP/CFI).
5. Validation & Delivery: Cross-check across methods; provide IOC set, exploitability verdict, mitigations, and archive artifacts at skills/security/reverse-engineering-extended/reverse-engineering-deep-analysis/{project}/{timestamp} with MCP tags (WHO=reverse-engineering-deep-analysis-{session}, WHY=skill-execution).

Deliverables

• Detailed analysis report, PoCs (where permitted), IOC inventory, and mitigation guidance.
• Evidence bundle (traces, dumps with hashes, configs, tool versions).

Quality Gates

• Structure-first documentation; missing resources/examples/tests tracked.
• Safety controls verified; dual validation for critical conclusions.
• Confidence ceilings stated for each claim; chain-of-custody recorded.

Anti-Patterns

• Skipping unpacking validation; trusting single-tool results.
• Executing with open network access.
• Claiming exploitability without demonstrated reach.

Output Format

• Scope + constraints table (HARD/SOFT/INFERRED).
• Findings, PoCs, and IOCs with evidence.
• Mitigations and validation log.
• Confidence line: Confidence: X.XX (ceiling: TYPE Y.YY) - reason.

Confidence: 0.72 (ceiling: inference 0.70) - Extended deep analysis aligned with skill-forge and prompt-architect standards.