PixCull is a single-developer project. The latest main is the only supported version. Patches for older commits are not backported.

If you think you've found a security issue, do not open a public issue. Email the maintainer at hello@pixcull.dev with:

• A short description of the issue
• A reproduction recipe (commands, payload, expected vs actual)
• The affected commit SHA from git rev-parse HEAD
• Your handle if you'd like credit in the fix commit

Expected turnaround: acknowledgement within 72 hours, fix or mitigation plan within 7 days for anything that lets a remote attacker read photos, exfiltrate annotations, or escalate within the host.

PixCull is designed to run locally on a photographer's machine or LAN deployment. It assumes:

• Trusted local user. The CLI + server APIs are not hardened against a malicious operator with shell access; treat the user as already authorized to read every photo in the configured roots.
• Localhost or controlled LAN. The server listens on 127.0.0.1 by default. Exposing it on a public network is supported only behind a reverse proxy + the X-PixCull-API-Key header check.
• Untrusted image input. EXIF + IPTC parsing goes through Pillow, which has had CVEs. We pin Pillow ≥ 10.2 and never run unsanitized metadata as code, but a malicious image is still potentially a resource-exhaustion vector.

Things that are explicitly out of scope:

• DoS via a giant batch (the user shipped that batch themselves)
• Self-XSS in the results page (filenames are HTML-escaped via esc(); if you find a path that isn't, that IS in scope — see V14.0)
• Reading the user's own files (the user invoked the tool)

Annotation files (annotations.jsonl) and score CSVs are designed to be portable across machines. They contain filenames, axis stars, free-form rationale text, and the cull_reason taxonomy token — but no raw images, no API keys, no license tokens. Sharing them with collaborators or attaching them to bug reports is safe.