- Sim facade (bound as `sim`): start(tag) arms the sim, dispatching 14a/15693 by tag
type; stop(), push(tag) -> tag.sync(), frames -> decoded trace
- SimResource: opens SimSession on load, closes on exit; a missing device is non-fatal
- one connection mode per workspace (reader OR sim) — the second is refused (serial contention)
- sim.push added to the write effect patterns (withheld on autoreplay)
Arming stays explicit. Hardware-free tests via injected sessions.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Exposes pm3py as a pyws workspace, discovered via the pyws.plugins entry point:
- ReaderResource: connects/disconnects a Proxmark3 as `reader`; a missing device is
non-fatal (reports disconnected)
- namespace injection of the transponder/sim classes, SimSession and Proxmark3
- effect declarations so autoreplay/replay withhold writes/field/sim/destructive while
reads and pure model edits replay
Hardware-free tests via AsyncMock. The live `sim` resource is P2.
See docs/plans/2026-07-05-pm3py-pyws-plugin-plan.md.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
build_firmware()/find_firmware_src() cross-compile fullimage.elf via
'make -C firmware PLATFORM=<explicit> armsrc/all' and pm3flash --build flashes it.
PLATFORM is explicit (never inferred from is_rdv4 — that's the running firmware's
flag, not the board).
Hardened per an adversarial multi-agent review of the change:
- --build now implies --force: it must flash the image it just built. A dirty
rebuild keeps the same base SHA, so matches_pin can't distinguish it from the
old firmware and would otherwise silently skip the flash (exit 0).
- make missing/not-executable -> FlashError (was an uncaught OSError traceback).
- find_firmware_src no longer falls back to a CWD-relative ./firmware — running
make on a stray Makefile is arbitrary code execution; only explicit arg /
$PM3PY_FIRMWARE_SRC / the package repo root. Suite 1045 pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Found on the first real-hardware flash (PM3 Easy). The fullimage ELF has two
adjacent PT_LOAD segments (seg1 ends at 0x15E5AC, seg2 starts there) that share
the 0x15E400 flash page. build_blocks aligned each segment independently, so the
shared page was written twice — seg2's 0xFF-padded head, written last,
erase-programmed the page and wiped seg1's real bytes there (the SAM7 programs a
whole page atomically). That region holds version_information, so the flashed OS
booted but reported 'Missing/Invalid version information'.
Fix: merge segments whose block-aligned spans touch into one buffer, so every
shared page is written once with both segments' real data; 0xFF only fills gaps
and aligned outer edges. Re-flash verified on hardware: 3df2ed2e8 -> 303bd5873,
valid version, matches_pin true. Bootrom untouched throughout (fullimage-only).
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Adds the 'is the device running the right firmware, and flash it if not'
layer on top of the flasher.
- _firmware.py: FirmwarePin (pinned to the firmware/ submodule commit) and
matches() — a device version-string test against the pinned git SHA (or a
release tag, once tagged releases exist).
- FirmwareInfo gains expected_firmware + matches_pin, computed on connect, so
pm3.firmware.matches_pin tells you whether the fork build is present.
- pm3flash console-script: auto-detect, compare to the pin, resolve the image
(--image / $PM3PY_FIRMWARE / local build), confirm, flash with progress,
then reopen and verify the new version. Flashes only on mismatch unless
--force; fullimage-only unless --allow-bootrom.
Detection + CLI are unit-tested; the auto-download layer (fetch the pinned
release asset) and hardware validation are still to come. Suite 1034 pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Drives the full flash flow over the existing USB-serial link — enter
bootloader, arm window, stream the ELF block-by-block, reset — with no
external pm3-flash/DFU/JTAG.
- core/flash.py: Flasher (device_info, auto OS->bootloader handover,
flash_image block loop with ACK/NACK+EFC status, reset, end-to-end flash);
ELF32-ARM PT_LOAD parsing; 0xFF-padded block alignment; image resolver.
- core/transport.py: OLD 544-byte frame encode/decode, send_old /
send_old_no_response, magic-branching frame reader, reopen() for USB
re-enumeration.
- core/protocol.py: DEVICE_INFO flags, START_FLASH_MAGIC, flash geometry,
BL version helpers.
- Wired as pm3.flasher; reachable through the sync proxy.
Safety: fullimage-only by default; bootrom region refused unless
allow_bootrom=True; every block bounds-checked against the flash window.
Verification is per-block ACK/NACK (as the C flasher does); read-back verify
and hardware validation are follow-ups. 22 tests; suite 1026 pass.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- NfcType2Tag._build_mfu_dump(): assemble the firmware mfu_dump_t (56-byte prefix
version/pages/signature + page data) so the sim serves real memory (UID/CC/NDEF/
PWD/PACK). Firmware is data-driven for tagtype 2/7, so an NTAG213 dump presents
correctly under tagtype 7.
- NfcType2Tag.sync(): push the image via CMD_HF_MIFARE_EML_MEMSET (same primitive
as the client's hf mfu esetblk), chunked. Updates a live sim with the fw host-poll.
- start_14a: load the dump before HF_ISO14443A_SIMULATE so the reader sees real
page data instead of an empty buffer (works on current fw; no reflash for this).
- fix: set_ndef/clear_ndef cleared memory to the end, wiping the lock/config/PWD/
PACK pages on NTAG parts. Bounded to the CC-declared NDEF area now.
- tests: mfu_dump layout, config-page preservation, EML_MEMSET framing/chunking.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
- start_14a now raises TypeError for a non-Tag14443A transponder, start_15693
likewise for a non-Tag15693 (local imports to avoid the base<->sim circular
import). Prevents silently starting a sim with an incompatible tag model.
- Re-export the full 14a/14b/15693 tag lineup (NTAG21x, Ultralight family, NTAG
I2C, ST25*, OPTIGA, ISO14443B, TI parts) from pm3py.sim for `from pm3py.sim
import NTAG213`-style use.
- Add a read-only `uid` property on the Transponder base (mirrors set_uid()).
- Tests for both guards.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
.gitignore had `venv/` but the project's virtualenv is `.venv/`; add it plus
local-only agent/editor state so a `git add -A` can't sweep them in. Also stop
tracking four .pyc files committed before the __pycache__ ignore rule existed.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Y1: Cmd.HF_ISO14443A_SIM_TRACE (0x0339).
Y2: generalize SimSession._trace_reader dispatch to a {cmd: protocol} map so
14a (proto 1) and 15693 (proto 0) share one reader; the ADC field-strength
branch is now guarded to 15693.
Y3: rewrite start_14a as a synchronous trace path mirroring start_15693 (raw
pyserial + the _trace_reader thread). Sends the correct HF_ISO14443A_SIMULATE
NG payload (same layout as hf.iso14a.sim()), sets FLAG_SIM_TRACE when
streaming, and takes trace/on_frame/quiet/tagtype/exit_after kwargs. Loading
tag memory into emulator RAM stays a separate (hardware) step; the trace
streams every reader<->tag frame regardless. The WTX _relay_loop is retained
for the future 14a-4 path.
T1: mocked-transport unit tests for the 14a trace path (tests/test_sim_trace_14a.py).
Requires the bumped firmware submodule. 996 tests pass.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
SniffSession/SimSession.start_15693 gain on_frame(frame) and quiet kwargs (defaults preserve today's print behavior). Each frame is a plain dict {protocol, direction, timestamp, duration, data, data_hex, decoded[, crc_fail]} with no ANSI. SimSession now accumulates frames (entries property + clear(), mirroring SniffSession) and auto-streams when trace or on_frame is set; printing is gated on trace and not quiet. TraceFormatter.decode() exposes the annotation without color; format()/print() untouched.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
hf.mfc renamed to hf.mf to match the C client's 'hf mf'. New hf.mfu (HFMFUCommands) provides rdbl/wrbl/dump over the dedicated CMD_HF_MIFAREU_READBL/WRITEBL using the firmware mful_readblock_t / mful_writeblock_t structs; a READ returns 4 pages (16 bytes) and the firmware replies them directly. keytype selects auth (0=none, 1=UL-C, 2=EV1/NTAG pwd, 3=UL-AES). Wire formats verified against firmware structs; needs a bench pass for the auth/write-length paths.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
inventory() parses both response formats: slots=1 -> single tag dict (or None); slots=16 -> list of tags from the iso15_inventory_response_t anti-collision struct, each CRC15-validated. Optional request-level AFI filter and client-side DSFID filter. One 16-slot round drops colliding slots (full tree-walk resolution is a follow-up); scan() stays the single-tag convenience.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
The sync REPL proxy forwarded methods via __getattr__, so dir()/tab-completion never saw them and the sync wrapper reported (*args, **kwargs) instead of the real signature. Add __dir__ delegating to the wrapped target, and decorate the async wrapper with functools.wraps so inspect.signature()/help() report the real params + docstring.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
lf.tune sends the 2-byte {mode,divisor} the firmware requires (was 3 -> EINVARG/0V). t55.writebl packs the 10-byte {data,pwd,blockno,flags} struct with a single bit-packed flags byte (was 12). t55.wakeup packs 5 bytes {pwd,flags}. t55.config is maintained client-side (firmware has no GET). LF_SAMPLING_SET_CONFIG uses send_ng_no_response (firmware never replies). Pre-existing, independent of the rebase.
14a.sim sent a MIX frame but the handler is NG casting a packed {tagtype,u16 flags,uid[10],exitAfter,rats[20],ulauth...} struct -> whole struct garbled, flags never arrived. Rewrite to send_ng with the exact struct and expose tagtype (2=Ultralight,7=NTAG) so Ultralight/NTAG sim works. 14a.sniff sent MIX (args stripped -> empty asBytes); send NG with the 1-byte param. Pre-existing, independent of rebase.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
dropfield waited on a reply the firmware never sends (bare hf_field_off();break;) -> 2s stall + raise; use send_ng_no_response. hf.sniff under-sent the 10-byte {u32,u32,u8,u8} struct and mis-parsed the uint16 reply. hw.standalone sent an empty payload for a {arg,mlen,mode[10]} struct and blocked; pack it + fire-and-forget. Pre-existing, independent of rebase.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
wrbl padded key to 16 (firmware reads block at asBytes+10) -> corrupt writes; pad to 10. nested sent {block,keytype,key,target...} but firmware struct is {block,keytype,target_block,target_keytype,calibrate,key[6]} -> attacked wrong target; reorder + add calibrate + decode reply. cident/sim sent empty/MIX payloads for NG handlers -> send the packed structs. mf.sniff (no firmware handler) reroutes to CMD_HF_ISO14443A_SNIFF. Pre-existing, independent of rebase.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ISO15_READ_RESPONSE was 0x10 but the firmware bit is 0x20 (0x10 is HIGH_SPEED, which pm3py lacked), so scan/rdbl/wrbl sent flags 0x19 with READ_RESPONSE never set -> firmware never listened and replied empty. The ARM firmware also ignores ISO15_APPEND_CRC (the C client adds the CRC host-side via AddCrc15), so pm3py's CRC-less frames were never answered by a real tag.
Fix flag values to match firmware include/iso15.h, add crc15 (CRC-16/X-25), and route all iso15 commands through _iso15_payload (appends CRC, sends CONNECT|HIGH_SPEED|READ_RESPONSE=0x31). Payload now byte-matches the C client getUID. Pre-existing bug, independent of the firmware rebase.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
CMD_MEASURE_ANTENNA_TUNING_HF requires a 1-byte mode arg (firmware rejects length!=1 with EINVARG) and returns the voltage as uint16, not uint32. tune() sent no arg (empty reply -> 0 mV) and parsed <I. Now sends mode 1/2/3 (START/MEASURE/STOP) and parses <H. Verified vs firmware appmain.c handler; C client reads the same ~15V on this firmware.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Merges the package refactor (core/ sub-package, normalized naming,
scaffolded sub-packages) while preserving all in-progress sniff
infrastructure. Git rename detection carried WIP changes to new paths.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
In-progress work — committing before refactor merge so git rename
detection can carry changes to new file locations.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Phase 1 of package refactor. Moves all source modules into pm3py/core/
with file renames (hf_14a→hf_iso14a, hf_15→hf_iso15, hf_mf→hf_mfc)
and client attribute normalization (hf.a14→hf.iso14a, hf.mf→hf.mfc).
pm3py/__init__.py re-exports from core for backward compat.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
