The script runs under #!/bin/sh; on dash the array assignment was a
syntax error. Pass the auth option and its value as two plain words.
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
* security: harden dev-tool scripts and the testkit skip-log test (CodeRabbit)
- tools/install-sam3.sh: the HF bearer token went through curl's argv, where
`ps` exposes it to other local accounts (CWE-214); it now travels through
a private 0600 temp config file removed on exit.
- packaging/macos/import-cert.sh: the decoded .p12 was written to a
predictable path in a shared TMPDIR, where a precreated attacker-readable
file would receive it (CWE-377); both cert and keychain are created inside
a mktemp -d 0700 directory now.
- docs/showcase/run.py: the reply reader copied whole responses with no size
cap (obvious quadratic-copy and unbounded-memory DoS). It accumulates in a
bytearray now and closes the connection past 128 MiB.
- tools/spectral_camera_table.py: alias/make/model strings and comments from
the rawtoaces-data checkout moved directly into Rust source; one newline in
a comment became Rust code. External text is now escaped as Rust literal
content and comments are normalized to one printable line.
- crates/testkit: appended_lines_accumulate reused a pip-predictable temp
dir by creating it with create_dir_all and carried on after failing to
remove its log path, so the append would follow an attacker-planted
symlink (CWE-59); it now fails the test if the directory exists instead of
testing through it, and removes any existing log path only after clearly
checking it, always failing the test instead of proceeding when removal fails.
Findings #10080/#10078 PR (rows 2, 6, 8, 9, 15 of the
storytold-lightcraft-agent-findings report).
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
* security: bounded catalog keyword, repair, journal recovery work (CodeRabbit)
- keywords: from_file searched every hierarchical level for each flat keyword
and `same` lower-cased two strings per comparison — one lowered level set
instead (CWE-407, row 97).
- keywords: parse_keyword_list scanned each keyword's growing synonym list per
line — a lowered-synonym set per keyword instead (CWE-407, row 100).
- keywords: keyword_tree stored every level's prefix twice, so a photo XMP
keyword nested deeper than a list file accepts made stored bytes quadratic
in the level count; the tree caps at MAX_LEVELS, and keyword_list_text names
whatever lies deeper from the photos' and list's own spellings, so the file
still reads back (CWE-400, row 103).
- catalog: repair validated a folder chain once per album — memoized by chain
now, so a valid chain in a flat snapshot costs one walk overall (CWE-407,
row 101).
- store: journal append and truncate follow a planted catalog.log symlink;
both refuse any non-regular file at the name before writing through it
(CWE-59, row 107), like the staging writes since #1502.
- journal: resync re-validated the whole remaining suffix UTF-8 for every
candidate record start; a whole line is validated once (covering every
candidate in it), and an invalid line's re-checks stop at a 64 MiB budget,
with the tail then staying a torn one the damaged path keeps and reports
(CWE-407, row 109).
New tests: from_file case-insensitive level match, synonyms kept once, a
4096-level keyword cut at MAX_LEVELS, a link at catalog.log refused, and
resync over a fragment of 1024 record starts.
Findings rows 97, 100, 101, 103, 107, 109 of the
storytold-lightcraft-agent-findings report.
🤖 Generated with Codebuff
Co-Authored-By: Codebuff <noreply@codebuff.com>
---------
Co-authored-by: Codebuff <noreply@codebuff.com>
* CI: dev profile keeps line tables for workspace crates and no debug info for dependencies
* Control server tests: persistent_requests_and_slow_replies_outlive_the_idle_read_timeout no longer races the idle timeout
Each request (and the closing half-shutdown) reaches the socket before the
worker reads, so on a slow machine only the reply waits outlast the 100 ms
idle timeout, as the test intends. The idle test sends its input the same
way, and the no-frames test allows 5 s instead of 2 s to give up.
* CI: one integration-test binary per crate
Each crate's tests/*.rs become modules of tests/it/main.rs (51 test files
into 13 binaries, plus 6 that stay their own). Inner #![cfg] attributes
move to the mod lines. Files that change state for the whole process keep
a binary of their own: codecs jxl_tail_heap and raw sr2_heap (dhat global
allocator), gpu shutdown, shutdown_first and memory, ui-egui shutdown.
Docs, comments and the GPU probe name the new paths and targets
(--test it <module>::); the CI report names an it binary's skips after
the crate.
* CI: tests under nextest, doctests apart, one wasm check
xtask ci runs the tests with cargo nextest run --lib --bins --tests when
cargo-nextest is installed, else cargo test with the same targets and a
one-line notice; doctests run after them with cargo test --workspace --doc.
The heif and hardware GPU steps use the same runner. Examples stay out of
the test build (clippy --all-targets still checks them). Concurrent tests
under nextest follow the RUST_TEST_THREADS cap (NEXTEST_TEST_THREADS).
The wasm check is one cargo check --keep-going over the same crates, and
clippy runs with --keep-going.
* CI: tests run under cargo test by default, nextest with LIGHTCRAFT_NEXTEST=1 (per-process scene cache)
Bash scripts use [[ ]] for tests, functions name their positional
parameters and end with an explicit return, and the app-specific packaging
check writes its ::error:: annotations to stderr. No behaviour change.
Resolve and validate command-line path arguments before use, enforce HTTPS at
download sinks, and confirm path containment. Hardens local codegen/fetch tools
against path-traversal and command-argument-injection findings.
Describe merged the detector's per-instance masks. Those stay near 0 even well inside what they
found: a sky came out about 87 % selected, a volleyball net about 70 %, a person about 94 %.
Two instances that touch both fade out where they meet, so a line showed between people standing
side by side.
- The detector's semantic head is now ported: semantic_projection over the pixel decoder, the
transformers implementation's semantic_seg. It matches the reference to about 1e-3, like the
instance masks. tools/sam3_reference.py writes it and the reference test checks it; older
reference files still pass.
- Each instance found (score above the threshold, as before) takes the semantic logits where it
is the strongest instance and near (above -4). The semantic mask is sure inside and runs on
across the boundary between two instances. Far from every instance nothing changes, so
Describe selects nothing it didn't find before.
On 16 prompts over three photos (a mountain landscape, two gym photos), the median coverage
inside each selection rises, e.g.:
- sky 0.87 -> 1.00
- mountains 0.81 -> 0.99
- net 0.71 -> 0.99
- floor 0.92 -> 1.00
- person 0.94 -> 0.99
Each selection gains 0 to 1.9 % of the photo at its soft rim and loses none.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A raw without colour matrices of its own now gets DNG ColorMatrix1/2 and
ForwardMatrix1/2 for its model when the model is one of the 52 cameras
in rawtoaces-data (Academy Software Foundation, Apache-2.0): 22 Canon,
11 Nikon, 11 Sony, 3 Fujifilm and 5 others, plus the other names its
alias list gives (Rebel T3i and Kiss X5 for the EOS 600D; bodies it
lists as sharing a sensor). tools/spectral_camera_table.py fits, under
Standard Illuminant A and D65, a white-preserving 3x3 to the measured
spectral sensitivities over rawtoaces-data's 190 training reflectances
and writes crates/raw/src/spectral_table.rs. No number comes from Adobe
or from a raw decoder. lightcraft_raw::spectral looks a camera up by
Exif make and model; the DNG colour model does the rest.
camera_preview::starting_colour takes the first available of: the
file's own matrices, a camera profile for the model (fitted to the
JPEG for tone), the spectral matrices, the file-local JPEG fit, the
neutral fallback. A profiled model whose profile fit fails takes its
spectral matrices. White balance stays relative to the as-shot look for
every raw without matrices of its own, so Photo::relative_wb and the
controls are unchanged; calibrate still pools covered models.
Tests: one per precedence step (synthetic), the table's lookup, aliases
and DNG shape, and a corpus test of the order on ILCE-7M4 (profile),
EOS 5D Mark III and D5100 (spectral), EOS 7D (own fit) and a DNG. The
ILCE-7RM2 and GFX100S corpus tests now expect the spectral matrices.
Attribution in assets/ATTRIBUTION.md and NOTICE. RENDER_CACHE_VERSION 23.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Engine (crates/engine/src/segment/):
- One worker thread owns the model and the encoded photo; the session
sends jobs (with the render job for the model input) and gets results
over channels, never locking what the model holds. In the app
(Segmenter::background) clicks, descriptions, prepares and detail
passes return at once and Session::segment_poll applies results (one
undo step per settled selection, pending clicks shown on the photo);
the CLI/MCP wait for them. Newer requests for the same selection
supersede queued ones.
- Every job runs under catch_unwind: a candle panic becomes an error and
the model is reloaded next time; busy counters are reset by drop
guards; a dead worker is restarted. The model unloads after 10 min idle.
- segment.model.status / .download {acknowledged: true} / .cancel: the
download runs on its own thread; status via atomics (try_lock only).
Without the model every AI request fails at once with "The SAM 3 model
is not installed…"; renders/exports never need it; `seg` params make
AI masks from stored segmentations without it.
- Hostile documents: at most 4 detail patches per component (capped on
deserialize and in the renderer), at most 64 Object clicks.
UI: Object / Describe without the model open "Download the SAM 3
Model?" (size, folder, SAM License link; Not Now / Download, progress,
cancel, Try Again with the reason, Continue once installed); progress in
the Masking panel when the dialog is closed. Reverted the PR's unrelated
default change (mask overlay off + hover-red) to main's behaviour.
tools/install-sam3.*: --fail so error pages never land in files, 416
restarts instead of looping, the weights are verified before they are
moved into place (a bad file is deleted, never resumed from).
Docs: docs/ai-masks.md (in-app download, mirrors, SAM License summary,
no weights in the repo), NOTICE, parity LR-MASK-OBJECTS, What's New,
ROADMAP.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
AI masks from Meta's SAM 3, run in LightCraft itself: a new
`lightcraft-segment` crate ports the Hugging Face transformers
implementation (Apache-2.0) to candle, on the GPU through Metal on macOS
(CPU elsewhere), with no C/C++ and no `unsafe`.
- Object: click to include, ⌥-click to leave a part out (SAM 3's
interactive prompt encoder + two-way mask decoder).
- Describe: a text prompt selects every instance ("sky", "the red car";
"car, road" merges several) via the CLIP text encoder and DETR detector.
- Both as new masks and as Add / Subtract / Intersect components; + / −
buttons under the selected mask combine selections.
- The segmentation is stored with the mask (288² quantized logits, ~10-30
KB), so renders, exports and the web build never need the model.
- A zoomed-in detail pass (background, ~5 s on an M4 Pro) re-segments the
area around a selection from a sharper source: 5-10x finer edges on
small objects.
- Per-selection Edge slider (hard ↔ soft); hovering a mask in the list
shows it in red (the always-on overlay is now off by default, O toggles).
- The engine feature `sam` (on in the desktop app) gates the model; the
image is encoded once per photo in the background, clicks take ~35 ms.
- Weights are not bundled (SAM License): tools/install-sam3.sh / .ps1
download and verify facebook/sam3; docs/ai-masks.md.
- Clearer errors (missing file, missing model) and a stderr logger in the
app (LIGHTCRAFT_LOG=info).
candle is pinned at 0.9.2: later releases make candle-core depend on
tokenizers with the Oniguruma C library. Every stage is checked against
the Python reference (tests/reference.rs, tools/sam3_reference.py): on
Metal the backbone matches to 7e-5 relative error, masks to ~5e-5.
Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01C8BkS7JDqa1EMcjW6spDGA
