Connect

Bundled with Unleash skills pack. Source: C:\Users\Admin.agents\skills\connect\SKILL.md

CLI Help (verbatim idasql --help)

idasql vX.Y.Z - SQL interface to IDA databases

Usage: idasql -s <file> [-q <query>] [-f <file>] [-i] [--export <file>]

Options:
  -s <file>            IDA database (.idb/.i64) OR raw binary (.exe/.dll/firmware/etc.)
                       — raw binaries trigger fresh idalib analysis and string-list rebuild
                       — legacy 32-bit .idb files upgrade to .i64 and require an explicit reopen
  --token <token>      Auth token for HTTP/MCP server mode (if server requires it)
  -q <sql>             Execute SQL query or semicolon-separated script
  -f <file>            Execute SQL from file
  -i                   Interactive REPL mode
  -w, --write          Save database on exit (persist changes)
  --export <file>      Export tables to SQL file (local mode only)
  --export-tables=X    Tables to export: * (all, default) or table1,table2,...
  --http [port]        Start HTTP REST server (default: 8080, local mode only)
  --bind <addr>        Bind address for HTTP/MCP server (default: 127.0.0.1)
  --mcp [port]         Start MCP server (default: random port, use in -i mode)
                       Or use .mcp start in interactive mode
  -h, --help           Show this help
  --version            Show version

Examples:
  idasql -s test.i64 -q "SELECT name, size FROM funcs LIMIT 10"
  idasql -s test.i64 -f queries.sql
  idasql -s test.i64 -i
  idasql -s test.i64 --export dump.sql
  idasql -s test.i64 --http 8080
  idasql -s sample.exe --http            # raw PE: idalib auto-analyzes, then serves SQL (default port 8080)
  idasql -s firmware.bin -q "SELECT * FROM binary"
  idasql -s test.i64 --mcp 9000

Key takeaway: -s accepts either an existing IDA database or a raw binary. No manual idat -A -B / ida -B pre-step is needed — point -s straight at the .exe/.dll/firmware/etc. and idalib creates the database on first open.

Additional Resources

• Canonical schema catalog: references/schema-catalog.md
• CLI reference, REPL commands, server modes, runtime controls: references/cli-reference.md (includes .pin autostart: the IDA plugin can auto-start a pinned HTTP/MCP server when a database is opened)

Autostart pinning (HTTP/MCP)

.pin http|mcp [bindinterface] [port] pins a server so the IDA plugin auto-starts it when the database opens (the pin lives in the IDB; CLI changes persist only with -w). The port is optional:

• Fixed port — .pin http 0.0.0.0 8080 → autostarts on that stable port every launch (use for stable automation).
• Random port — .pin http (port omitted, stored as 0) → autostarts on a fresh random port each launch (a real pinned setting, not "unset"); read the chosen port from the live status. .pin status shows <random port each launch>.

[bindinterface] is the host/interface (default 127.0.0.1). .pin on|off http|mcp toggles autostart without dropping the host/port; .pin clear [http|mcp|all] removes it. .http start / .mcp start with no port reuse the pinned host/port. Full grammar: references/cli-reference.md.

• Optimization quality gate: references/optimization-checklist.md
• HTTP server guide: references/server-guide.md

Quick Start Guardrails

• Always provide -s <file>. The file can be either an IDA database (.idb / .i64) or a raw binary (.exe, .dll, firmware blob, etc.) — raw binaries trigger fresh idalib auto-analysis and string-list rebuild. Do not pre-create an IDB with idat -A -B; idasql -s raw.exe does it in one shot.
• If idasql -s legacy.idb ... exits with code 3 and stdout JSON containing {"status":"upgraded","reopen_with":"..."}, repeat the same requested operation with -s <reopen_with>. Server modes exit before binding in this case.
• Use --write when you want edits persisted on exit, including HTTP/MCP server shutdown.
• SQL inputs to -q, HTTP /query, MCP idasql_query, and plugin CLI may be one statement or a semicolon-separated script. Semicolons inside quoted strings are safe.
• All /query responses use the canonical script envelope — single statement = array of one: {success, statement_count, results:[{statement_index, success, columns, rows, row_count, elapsed_ms, error}], row_count_total, elapsed_ms_total, first_error_index}. Fail-fast is the default. Only the REPL/plugin .http server honors ?continue_on_error=1 in the query string to run every statement regardless of earlier failures; the CLI --http handler ignores it and MCP has no such argument.
• Discover schema before writing queries:
  • REPL: .schema <table>
  • SQL: PRAGMA table_xinfo(<table>);
• Start orientation with SELECT * FROM binary;.

Schema Catalog

Canonical column shapes and owner-skill mapping live in references/schema-catalog.md. Sourced from pragma_table_list + pragma_table_xinfo. Use it before assuming column names for less-common surfaces.

Session Bootstrap Contract

Use this exact startup flow before deep analysis:

1. Connect to database (-s, -i, or --http). -s accepts either an existing IDB (.idb/.i64) or a raw binary (.exe/.dll/firmware/etc.) — never pre-build an IDB with idat; let idasql -s raw.exe do it. If a legacy .idb upgrades and returns status:"upgraded", restart with the JSON reopen_with path before running orientation.
2. Run orientation query:

SELECT * FROM binary;

3. Validate key entities exist:

SELECT COUNT(*) AS funcs FROM funcs;
SELECT COUNT(*) AS xrefs FROM xrefs;
SELECT COUNT(*) AS strings FROM strings;

4. Introspect schema for target surfaces before authoring complex SQL:

PRAGMA table_xinfo(funcs);
PRAGMA table_xinfo(xrefs);

5. Route to domain skill using routing matrix below.

Global Agent Contracts

These contracts apply across all idasql skills and should be treated as one shared agent behavior model.

Read-First Contract

• Read current state first (SELECT) before writes (INSERT/UPDATE/DELETE).
• Confirm target precision using stable identifiers (addr, func_addr, idx, label_num).

Anti-Guessing Contract

• Do not assume columns/types for long-tail surfaces.
• Introspect via .schema or PRAGMA table_xinfo(...) before issuing uncertain queries.

Mandatory Mutation Loop

1. Read current state.
2. Apply mutation.
3. Refresh if needed (decompile(..., 1) for decompiler surfaces).
4. Re-read and verify expected change.

Performance Contract

• Always constrain high-cost surfaces (xrefs, instructions, ctree*, pseudocode) by key columns.
• For decompiler surfaces, enforce func_addr = X unless explicitly asked for broad scans.
• For raw dirtree browsing, prefer tree = ? plus path, path LIKE, or parent_path; for normal organization use funcs.folder_path and types.folder_path.
• dirtree_entries is read-only diagnostics. Recursive folder delete and raw recovery/link operations are intentionally not SQL surfaces.

Failure Recovery Contract

• On no such table/column: introspect schema and retry.
• On empty results: validate address range, table freshness (rebuild_strings()), and runtime capabilities.
• On timeout: narrow scope, add constraints, paginate, or split query.

Output Contract

• Selection - decide whether and how much to surface from user intent. Answer questions directly ("biggest is main, 500 bytes"); show supporting rows only when they help the user verify; don't dump full tables unprompted; never surface data fetched only as an intermediate reasoning step.
• Fidelity - when you do present code/data, show the real artifact (decompilation, actual rows), never a paraphrase.
• Mechanics - the HTTP /query response is a JSON envelope ({success, results:[{columns,rows,...}]}). Consume it directly and render in your reply. Do not pipe responses through python/jq to pre-render a table - that discards the success/elapsed_ms/error fields and makes you reason over a lossy view. The CLI (-q/-f) already prints a table. Reserve jq/python for extracting a value to feed a later query. (For direct terminal/pipe use the server can emit ?format=text|csv|tsv; as an agent, consume json.)

Skill Routing Matrix (Intent -> Skill)

Use this deterministic mapping for initial routing:

When prompts span domains, execute in this order:

1. Orientation in connect
2. Primary domain skill
3. Adjacent skills for enrichment (for example xrefs + decompiler + annotations)

Cross-Skill Execution Recipes

Recipe: Unknown binary triage -> suspicious function deep dive -> annotate

1. analysis: identify candidates from imports/strings/call patterns.
2. xrefs/disassembly: map call graph and call sites.
3. decompiler: inspect logic and variable semantics.
4. annotations: apply comments/renames/types with mutation loop.

Recipe: String IOC -> reference graph -> patch

1. data: locate candidate strings and addresses.
2. xrefs: map references to caller functions.
3. debugger or annotations: patch or annotate specific sites.

Recipe: Type recovery from pseudocode

1. decompiler: inspect lvars, call args, and ctree patterns.
2. types: create/refine structs/enums and apply declarations.
3. annotations: finalize naming/comments and verify rendered pseudocode.

UI Context Routing

For prompts like "what am I looking at?", "what's selected?", "what is on the screen?", "look at what I'm doing", or references to "this/current/that", use the dedicated ui-context skill.

ui-context owns:

• get_ui_context_json() capture/reuse policy
• temporal reference rules (this vs that)
• response template, examples, and fallback messaging

Runtime caveat:

• get_ui_context_json() is plugin GUI runtime only, not idalib/CLI mode.
• If unavailable, state that UI context is unavailable and continue with non-UI SQL workflows.

binary

Database orientation surface for quick session metadata. This is metadata-only and not a replacement for UI context capture.

Key/value shape — (key TEXT, value TEXT, type TEXT), one row per metadata fact, type ∈ {string, hex, bool, int}. The shape and the canonical key names are shared across the tool family, so the same WHERE key = '…' query works on every engine. The summary row is emitted first, so SELECT * FROM binary shows the digest up top.

Use the filename, idb_path, md5, or sha256 keys to confirm which binary/instance this connection is bound to.

SELECT * FROM binary;
SELECT key, value FROM binary WHERE key IN ('filename','idb_path','md5','sha256');
SELECT value FROM binary WHERE key = 'idasql_version';

For canonical schema and owner mapping, see references/schema-catalog.md.

What is IDA and Why SQL?

IDA Pro is the industry-standard disassembler and reverse engineering tool. It analyzes compiled binaries (executables, DLLs, firmware) and produces:

• Disassembly - Human-readable assembly code
• Functions - Detected code boundaries with names
• Cross-references - Who calls what, who references what data
• Types - Structures, enums, function prototypes
• Decompilation - C-like pseudocode (with Hex-Rays plugin)

IDASQL exposes all this analysis data through SQL virtual tables, enabling:

• Complex queries across multiple data types (JOINs)
• Aggregations and statistics (COUNT, GROUP BY)
• Pattern detection across the entire binary
• Scriptable analysis without writing IDA plugins or IDAPython scripts

Core Concepts for Binary Analysis

Addresses (ea_t)

Everything in a binary has an address - a memory location where code or data lives. IDA uses ea_t (effective address) as unsigned 64-bit integers. SQL shows these as integers; use printf('0x%X', addr) for hex display.

Address-taking SQL functions accept:

• integer EA values (preferred for deterministic scripts)
• numeric strings ('4198400', '0x401000')
• symbol names resolved with get_name_ea(BADADDR, name) (global names)

Most table predicates should compare address columns to integer EAs such as addr = 0x401000. applied_types.addr is an intentional exception for equality filters/writes: it also accepts numeric strings and symbol names so type application can target names directly.

Examples:

SELECT decompile('DriverEntry');
UPDATE applied_types
SET decl = 'NTSTATUS DriverEntry(PDRIVER_OBJECT, PUNICODE_STRING);'
WHERE addr = 'DriverEntry';
SELECT (SELECT comment FROM comments WHERE addr = 0x401000 LIMIT 1);

Read address comments from the comments table after resolving the target EA.

If a symbol cannot be resolved, SQL functions return an explicit error like: Could not resolve name to address: <name>.

Local label lookup that depends on a specific from context is not consulted by default (BADADDR resolution). Use explicit numeric EAs when needed.

Functions

IDA groups code into functions with:

• addr / start_addr - Where the function begins
• end_addr - Where it ends
• name - Assigned or auto-generated name (e.g., main, sub_401000)
• size - Total bytes in the function

There will be addresses and disassembly listing not belonging to a function. IDASQL can still get the bytes, disassembly listing ranges, etc. For single-EA disassembly (code or data), prefer disasm_at(addr[, context]) over function-scoped queries.

Cross-References (xrefs)

Binary analysis is about understanding relationships:

• Code xrefs - Function calls, jumps between code
• Data xrefs - Code reading/writing data locations, or data referring to other data (pointers)
• from_addr -> to_addr represents "address X references address Y" Use table: xrefs(from_addr, to_addr, type, is_code).

Segments

Use table: segments(start_addr, end_addr, name, class, perm) (full CRUD). UPDATE start_addr rebases a segment, end_addr resizes it.

Memory is divided into segments with different purposes. For example, a typical PE file, has these segments:

• .text - Executable code (typically)
• .data - Initialized global data
• .rdata - Read-only data (strings, constants)
• .bss - Uninitialized data

Of course, segment names and types can vary. You may query the segments table to understand memory layout.

Basic Blocks

Within a function, basic blocks are straight-line code sequences:

• No branches in the middle
• Single entry, single exit
• Useful for control flow analysis Use table: blocks(start_addr, end_addr, func_addr, size).

Decompilation (Hex-Rays)

The Hex-Rays decompiler converts assembly to C-like pseudocode:

• ctree - The Abstract Syntax Tree of decompiled code
• lvars - Local variables detected by the decompiler
• Much easier to analyze than raw assembly

Core decompiler surfaces:

• decompile(addr) (PRIMARY read/display surface)
  • Returns the entire function as one text block.
  • Each output line is prefixed for address grounding:
    • Addressed line: /* 401010 */ ...
    • Non-anchored line: /* */ ... (no address anchor for that line)
  • Use this first when the user asks to "decompile", "show code", "show pseudocode", or "explain function logic".
• pseudocode table (structured/edit surface)
  • Use for line-level filtering (func_addr, addr, line_num) and comment writes keyed by addr + comment_placement.
  • Resolve a writable pseudocode anchor first; do not assume addr == func_addr.
  • Not the preferred display surface for full-function code.
• ctree and ctree_call_args for AST-level analysis
• ctree_lvars for local variable rename/type/comment updates

Performance Rules

CRITICAL: Constraint Pushdown

Some tables have optimized filters that use efficient IDA SDK APIs:

Always filter decompiler tables by func_addr!

Use Integer Comparisons

-- SLOW: String comparison
WHERE mnemonic = 'call'

-- FAST: Integer comparison
WHERE itype IN (16, 18)  -- x86 call opcodes

O(1) Random Access

-- SLOW: O(n) - sorts all rows
SELECT addr FROM funcs ORDER BY RANDOM() LIMIT 1;

-- FAST: O(1) - direct index access
SELECT addr
FROM funcs
WHERE rowid = ABS(RANDOM()) % (SELECT COUNT(*) FROM funcs);

CTE-First Mutation Workflow

For instruction lifecycle edits, use a CTE to identify precise targets first, then mutate:

WITH target AS (
    SELECT addr
    FROM instructions
    WHERE func_addr = 0x401000
    ORDER BY addr DESC
    LIMIT 1
)
DELETE FROM instructions
WHERE addr IN (SELECT addr FROM target);

SELECT make_code_range(addr, end_addr) FROM funcs WHERE addr = 0x401000;

This keeps mutation scope explicit and predictable for both humans and agents.

Summary: When to Use What

Remember: Always use func_addr = X constraints on instruction and decompiler tables for acceptable performance.

Error Handling

• No Hex-Rays license: Decompiler tables (pseudocode, ctree*, ctree_lvars) will be empty or unavailable
• No constraint on decompiler tables: Query will be extremely slow (decompiles all functions)
• Invalid address: Containing-function table lookups return no row; use a scalar subquery when you need a nullable scalar result
• Missing function: JOINs may return fewer rows than expected