If you discover a security vulnerability in kittyview, please report it through GitHub's private vulnerability reporting.

Do not open a public issue for security vulnerabilities.

Security-relevant areas of kittyview include:

• SVG rendering: SVGs are a rich format that can reference external files, embed scripts, and contain deeply nested structures. kittyview uses resvg (pure Rust, no scripting support) and defaults to blocking external file access (--svg-resources none).
• SVG foreignObject conversion: SVGs containing <foreignObject> elements (common in mermaid-cli, draw.io, and D3.js output) are preprocessed before rendering. Embedded HTML is stripped to plain text and replaced with native SVG <text> elements. No HTML is interpreted or executed -- all markup is discarded and only text content is preserved. Entity decoding is limited to a fixed set of named entities and numeric character references.
• Image decoding: Malformed images could trigger bugs in decoder libraries. All decoders are pure Rust (no C code).
• Terminal escape sequences: Malformed output could corrupt terminal state. kittyview buffers all protocol output before writing and validates terminal support before emitting. kittyview also reads from the terminal -- capability query replies, geometry reports, the slideshow's transmission checks, and keyboard input in raw mode. Everything read is treated as untrusted: reads are time-bounded, and replies and keys go through pure-Rust slice parsers that discard what they do not recognise (the key decoder additionally caps how many bytes one sequence may consume).
• Unsafe code: every unsafe block in kittyview is terminal-control FFI to libc -- termios raw mode, select(2), TIOCGWINSZ, reads and writes on terminal file descriptors, and the slideshow's signal handling (sigaction, kill) -- or zero-initialisation of the C structs those calls fill. Signal handlers do nothing but store an atomic flag, keeping them async-signal-safe. No unsafe code touches untrusted data: images, SVGs, and terminal replies are parsed entirely in safe Rust, and every hand-written parser facing untrusted input is fuzzed (see fuzz/): the PNG and foreignObject input-file parsers, the terminal-reply parsers, and the slideshow's key decoder.

Release binaries are built and attested with SLSA build provenance by the release workflow. To verify that an asset came from the release it claims to -- see Verifying downloads for why --source-ref is what gives a passing check that meaning:

gh attestation verify kittyview-linux-amd64.tar.gz \
  --repo gominimal/kittyview \
  --source-ref refs/tags/v0.1.7 \
  --deny-self-hosted-runners

Substitute the tag of the release you downloaded.

Each release also includes the provenance bundle itself (kittyview-provenance.intoto.jsonl) as an asset, so verification does not have to query GitHub's attestation store. Fully offline verification also needs a copy of the Sigstore trusted root, saved while still online:

gh attestation trusted-root > trusted_root.jsonl

gh attestation verify kittyview-linux-amd64.tar.gz \
  --repo gominimal/kittyview \
  --source-ref refs/tags/v0.1.7 \
  --deny-self-hosted-runners \
  --bundle kittyview-provenance.intoto.jsonl \
  --custom-trusted-root trusted_root.jsonl