About debugger
debugger is a skill for IDA debugging workflows, intended for analysts managing breakpoints, watchpoints, and byte patches. It documents breakpoint and patch inventories, mutation and recovery steps, and examples of setting breakpoints from analysis findings.
Key features
- 01Create, update, and delete breakpoints
- 02Manage conditional breakpoints and hardware watchpoints
- 03Read and modify mapped bytes
- 04Inventory patches and restore original bytes
- 05Set breakpoints in batches from call sites
Use cases
- 01Set breakpoints at calls to security-sensitive APIs
- 02Watch writes to structure fields
- 03Record and inspect binary patches
- 04Revert patches or verify modifications
Install
Copy the command and run it in a terminal.
npx skills add https:/ / github.com/ allthingsida/ idasql-skills --skill debuggerAfter installing, type/debuggerin Claude Code to confirm it loaded.
Bundled filesagents/openai.yaml
Details
From the project READMETrigger Intents
Use this skill when user asks to:
- add/remove/modify breakpoints
- patch bytes or revert patches
- create patch inventories and debugging action plans
- instrument analysis-driven break/watch workflows
Route to:
analysis/xrefsfor selecting meaningful targets firstdisassemblyfor opcode-level patch contextannotationsfor documenting patch rationale and outcomes
Do This First (Warm-Start Sequence)
-- 1) Current breakpoint inventory
SELECT printf('0x%X', addr) AS addr, type_name, enabled
FROM breakpoints
ORDER BY addr;
-- 2) Current patch inventory (fast: backed by IDA's patch list)
SELECT printf('0x%X', addr) AS addr, original_value, value AS patched_value
FROM bytes WHERE is_patched = 1
ORDER BY addr
LIMIT 50;
-- 3) Validate target bytes before patch
SELECT addr, value, original_value, is_patched
FROM bytes
WHERE addr = 0x401000;
Interpretation guidance:
- Confirm existing instrumentation before adding more.
- Always snapshot current/original byte state before mutating.
Failure and Recovery
- Breakpoint insert/update failed:
- Validate address existence and hardware size/type compatibility.
- Patch verification mismatch:
- Re-read
bytes(andWHERE is_patched = 1), then retry with precise address.
- Re-read
- Unintended patch side effects:
- Revert with
DELETE FROM bytes WHERE addr = ...and reassess target instruction context.
- Revert with
Handoff Patterns
debugger->disassemblyto validate instruction semantics around patch site.debugger->xrefsto assess blast radius of patched/broken call paths.debugger->annotationsto leave durable analyst breadcrumbs.
breakpoints
Debugger breakpoints. Supports full CRUD (SELECT, INSERT, UPDATE, DELETE). Breakpoints persist in the IDB even without an active debugger session. folder_path is the folder-oriented spelling of IDA's breakpoint group; updating either updates the same underlying breakpoint grouping.
| Column | Type | RW | Description |
|---|---|---|---|
addr | INT | R | Breakpoint address |
enabled | INT | RW | 1=enabled, 0=disabled |
type | INT | RW | Breakpoint type (1=hw_write, 2=hw_read, 3=hw_rdwr, 4=software, 8=hw_exec) |
type_name | TEXT | R | Type name (software, hardware_write, etc.) |
size | INT | RW | Breakpoint size (for hardware breakpoints) |
flags | INT | RW | Breakpoint flags |
pass_count | INT | RW | Pass count before trigger |
condition | TEXT | RW | Condition expression |
loc_type | INT | R | Location type code |
loc_type_name | TEXT | R | Location type (absolute, relative, symbolic, source) |
module | TEXT | R | Module path (relative breakpoints) |
symbol | TEXT | R | Symbol name (symbolic breakpoints) |
offset | INT | R | Offset (relative/symbolic) |
source_file | TEXT | R | Source file (source breakpoints) |
source_line | INT | R | Source line number |
is_hardware | INT | R | 1=hardware breakpoint |
is_active | INT | R | 1=currently active |
group | TEXT | RW | Breakpoint group name |
bptid | INT | R | Breakpoint ID |
folder_path | TEXT | RW | Alias for breakpoint group folder; NULL means root |
full_path | TEXT | R | Full breakpoint dirtree path when available |
-- List all breakpoints
SELECT printf('0x%08X', addr) as addr, type_name, enabled, condition
FROM breakpoints;
-- Add software breakpoint
INSERT INTO breakpoints (addr) VALUES (0x401000);
-- Add hardware write watchpoint
INSERT INTO breakpoints (addr, type, size) VALUES (0x402000, 1, 4);
-- Add conditional breakpoint (at a different address than the plain one above)
INSERT INTO breakpoints (addr, condition) VALUES (0x401330, 'eax == 0');
-- Move breakpoint into/out of an IDA breakpoint folder
UPDATE breakpoints SET folder_path = 'idasql/breakpoints/network' WHERE addr = 0x401000;
UPDATE breakpoints SET folder_path = NULL WHERE addr = 0x401000;
-- Disable a breakpoint
UPDATE breakpoints SET enabled = 0 WHERE addr = 0x401000;
-- Delete a breakpoint
DELETE FROM breakpoints WHERE addr = 0x401000;
-- Find which functions have breakpoints
SELECT b.addr, f.name, b.type_name, b.enabled
FROM breakpoints b
JOIN funcs f ON b.addr >= f.addr AND b.addr < f.end_addr;
bytes (Byte Patching)
Pure mapped-byte program view with patch support. This table is one row per
mapped byte address; IDA item metadata such as size/type belongs to heads.
| Column | Type | RW | Description |
|---|---|---|---|
addr | INT | R | Byte address |
value | INT | RW | Current byte value (UPDATE patches 1 byte) |
word | INT | RW | 2-byte little-endian value (UPDATE patches 2 bytes) |
dword | INT | RW | 4-byte little-endian value (UPDATE patches 4 bytes) |
qword | INT | RW | 8-byte little-endian value (UPDATE patches 8 bytes) |
original_value | INT | R | Original byte value before patch |
is_patched | INT | R | 1 if byte differs from original (WHERE is_patched = 1 enumerates patches fast) |
fpos | INT | R | Physical/input file offset (NULL when unavailable) |
Revert a patch with DELETE FROM bytes WHERE addr = ... (or WHERE is_patched = 1).
-- Read one address
SELECT addr, value, original_value, is_patched
FROM bytes WHERE addr = 0x401000;
-- Read a byte range, including item-tail bytes
SELECT addr, value
FROM bytes
WHERE addr >= 0x401000 AND addr < 0x401010
ORDER BY addr;
-- Get item metadata separately
SELECT addr, size, type, flags, disasm
FROM heads
WHERE addr = 0x401000;
-- Patch via table update (single byte, or width columns word/dword/qword)
UPDATE bytes SET value = 0x90 WHERE addr = 0x401000;
UPDATE bytes SET dword = 0x90909090 WHERE addr = 0x401000; -- little-endian
-- Inspect patch inventory (fast: backed by IDA's patch list)
SELECT printf('0x%X', addr) AS addr, original_value, value AS patched_value
FROM bytes WHERE is_patched = 1 ORDER BY addr LIMIT 20;
-- Persist once done
SELECT save_database();
bytes — Patching (write surface)
All byte patching is done through the writable bytes table.
| Column | RW | Meaning |
|---|---|---|
value | RW | Current byte; UPDATE patches 1 byte |
word / dword / qword | RW | UPDATE patches 2/4/8 bytes little-endian |
original_value | R | Original (pre-patch) byte |
is_patched | R | 1 if patched; WHERE is_patched = 1 enumerates patches fast |
SELECT printf('0x%X', addr) AS addr,
printf('0x%02X', original_value) AS old,
printf('0x%02X', value) AS new
FROM bytes WHERE is_patched = 1
ORDER BY addr;
Byte Access via the bytes Table
All byte reads and patches go through the bytes table. Bounded-read
shapes are documented in data; this skill focuses on the patching
workflow. load_file_bytes(...) writes a file's bytes into the IDB at
a given range; use it when the patch content already lives in a file.
| Function / Shape | Description |
|---|---|
SELECT hex(blob_concat(value)) FROM (SELECT value FROM bytes WHERE start_addr = X AND n = N ORDER BY addr) | Read N bytes as uppercase hex |
SELECT blob_concat(value) FROM (SELECT value FROM bytes WHERE start_addr = X AND n = N ORDER BY addr) | Read N bytes as BLOB |
load_file_bytes(path, file_offset, addr, size[, patchable]) | Write a file's bytes into the IDB at the target range |
-- Read 16 bytes as hex
SELECT hex(blob_concat(value))
FROM (SELECT value FROM bytes WHERE start_addr = 0x401000 AND n = 16 ORDER BY addr);
-- Patch one byte (example: NOP) and a 4-byte little-endian value
UPDATE bytes SET value = 0x90 WHERE addr = 0x401000;
UPDATE bytes SET dword = 0x90909090 WHERE addr = 0x401000;
-- Verify current vs original
SELECT value AS current, original_value AS original
FROM bytes WHERE addr = 0x401000;
-- Revert patch (one byte, or every patch)
DELETE FROM bytes WHERE addr = 0x401000;
DELETE FROM bytes WHERE is_patched = 1;
-- Persist patches explicitly
SELECT save_database();
Analysis-Driven Breakpoint Workflows
Set breakpoints on all callers of a security-sensitive API
Use disasm_calls to find every call site and batch-insert breakpoints:
-- Breakpoint on every call to VirtualAlloc (or similar)
INSERT INTO breakpoints (addr)
SELECT addr FROM disasm_calls WHERE callee_name LIKE '%VirtualAlloc%';
-- Verify
SELECT printf('0x%08X', addr) AS addr, type_name, enabled
FROM breakpoints;
Watchpoints on struct fields discovered via type analysis
After recovering a struct, set hardware watchpoints on specific field offsets:
-- Hardware write watchpoint on a 4-byte field (e.g., config.flags at base+0x10)
-- First, find where the struct base is stored (requires manual analysis)
INSERT INTO breakpoints (addr, type, size) VALUES (0x402010, 1, 4);
-- type=1 is hardware_write, size=4 for DWORD field
Conditional breakpoints from decompiler analysis
Set breakpoints that only trigger when specific conditions are met:
-- Break when first argument (rcx on x64 fastcall) equals a specific enum value
INSERT INTO breakpoints (addr, condition)
VALUES (0x401000, 'rcx == 3');
-- Break on error return
INSERT INTO breakpoints (addr, condition)
VALUES (0x401050, 'rax == 0xFFFFFFFF');
Patching Workflows
NOP out anti-debug checks
Find and neutralize IsDebuggerPresent checks:
-- Find calls to IsDebuggerPresent
SELECT dc.addr, (SELECT name FROM funcs WHERE dc.func_addr >= addr AND dc.func_addr < end_addr LIMIT 1) AS func_name,
disasm_at(dc.addr, 2) AS context
FROM disasm_calls dc
WHERE dc.callee_name LIKE '%IsDebuggerPresent%';
-- Patch the conditional jump after the check (example: jnz → nop nop)
-- First inspect the instruction after the call
SELECT disasm_at(0x401030, 3);
-- Then patch (adjust addresses based on actual binary)
UPDATE bytes SET value = 0x90 WHERE addr = 0x401035;
UPDATE bytes SET value = 0x90 WHERE addr = 0x401036;
Inventory all patches and generate report
-- Full patch report: what was changed and where
SELECT printf('0x%X', addr) AS addr,
(SELECT name FROM funcs WHERE addr >= addr AND addr < end_addr LIMIT 1) AS func_name,
printf('0x%02X', original_value) AS original,
printf('0x%02X', value) AS patched,
disasm_at(addr) AS context
FROM bytes WHERE is_patched = 1
ORDER BY addr;
Performance Notes
| Table | Size | Constraint | Notes |
|---|---|---|---|
breakpoints | Small (<100 typical) | none needed | Always fast |
bytes | All mapped bytes | addr | Critical — constrain to one address or a tight range |
bytes WHERE is_patched = 1 | Small (patch count) | is_patched = 1 | Fast — iterates only patched locations via IDA's patch list |
breakpointstable is small — full scans are fine.bytestable emits one row per mapped byte. UseWHERE addr = Xor a tightaddrrange.bytes WHERE is_patched = 1iterates only patched locations — always fast.
See Also
disassembly— instruction context around a patch site (disasm_at, operand semantics).annotations— record patch rationale via comments/bookmarks before mutating.data— canonical owner of byte-read shapes (table vs. scalar disambiguation).