The C-APDU TLV length was written as a raw byte: a 245-byte LOAD command
produced 'AE 80 22 F5 ...', which BER reads as a long-form marker, so the
card mis-parsed every script command over 127 bytes. Small commands worked,
which made a RAM install look alive: the card answered the LOAD steps with a
degenerate 'AF 80' body (no R-APDU), and the final INSTALL [for install]
failed with 6A88 because the package never loaded.
Both the indefinite ('22' TLV) and definite ('AA' outer) lengths are now
BER-encoded (same rule as the BIP channel data TLV fix). Tests: 245-byte
LOAD body, short form, definite variant (200 python, 346 frontend);
findings doc updated; service worker v148.
- shared _cap_apdu_sequence helper (INSTALL [for load] -> 240-byte LOAD
blocks -> INSTALL [for install]); the SCP80 /api/ram-install path now uses
it too (one source of truth; byte-level tests pin the APDUs)
- POST /api/scp81/ram-install: parses the .cap server-side and queues the
APDU sequence as the SCP81 command script (one C-APDU per POST, runs on the
card's next push); refused while a script is mid-run unless force
- /api/scp81/script reports the script kind (explore/none/custom/ram-install)
- tab: RAM install row (CAP file + SD AID + Queue button); RU strings
- docs: api.md, findings, AGENTS; service worker v145
Long GET STATUS listings answer SW CAFE with 127-byte pages. The script
responder now extracts the last complete AID from the page and inserts a
next-occurrence GET STATUS (80F2 <P1> 02 <Lc> 4F <len> <AID> 00) as the
next command, until the listing ends. Pages are logged (script-page), a
repeated page logs script-page-stalled and stops, inserted continuations are
dropped at session start.
Live-verified: ELF registry and applications collected completely in two
pages each (7/7 commands, all script-status ok).
- tests: page parsing (truncated tails, live FC-prefixed junk), continuation
bytes, auto-insert, stall guard, per-session purge (192 python + 337 frontend)
- UI: page number and script-status log rendering; service worker v143
- scp81.py: PSK TLS listener (stdlib ssl PSK callbacks) speaking the GP
HTTP administration dialog; configurable framing (chunked/Content-Length,
TLS record split, Apache-style/compact headers, Connection header,
keep-alive, Next-URI template with %d, TLS version/cipher, answer delay,
keylog for capture decryption)
- server.py: script responder + Response Scripting parsing (AF/AB, 80/23
TLVs), memory decoder, SCP81 start options, terminal-side timer
management, background-mode BIP events, permissive OPEN CHANNEL
- BIP fix: the RECEIVE DATA channel-data TLV length is BER long form
(36 81 <len>) above 127 bytes; a raw length byte is mis-parsed on the
card, so the large TLS records never reached its stack (a live card
fetched the script response and silently never processed it - endless
resume). The card now executes scripts and returns R-APDUs: memory
(13 applets, 50646 B NV free, 2402 B volatile), ISD, stored HTTP OTA
parameters, ELF and application registries
- frontend: SCP81 tab (listener, script selection, HTTP OTA log), phone
event forms, i18n; service worker v141
- docs: api.md, scp81-findings.md (attempt matrix + root cause analysis);
tools/scp81_decrypt.py decrypts listener captures via the keylog
- tests: 187 python + 337 frontend
catarrh