Changelog

TODO: wire this to the Debian changelog or GitHub releases so it auto-updates. For now, see the GitHub releases page or debian/changelog for the canonical version history.

v3.4.0 — fread media reading + CAVE docs site

Added image and PDF reading to fread, replacing the binary-skip path with a Python engine (fread-media.py) that handles PNG/JPEG/GIF/WEBP and PDFs via PyMuPDF (Poppler fallback).

Added

• New flags: —render, —pages, —meta-only, —no-resize, —max-pages, —max-tokens (image-specific), —token-budget (PDF text total cap), —no-ingest
• MCP fread emits image content blocks (mimeType per 2025-11-25 spec) for image and PDF render outputs; bypasses cli() for media payloads
• mcp/memory-ingest.js — detached, timed (3s) Node helper that writes ingest_payload to ShieldCortex via MCP stdio; best-effort, never blocks fread; supports recall-based dedupe by content hash
• Top-level "media_payload" field in fread JSON output for downstream parsers
• Error codes: PDF_ENCRYPTED, INVALID_PAGE_RANGE, BACKEND_MISSING, TOKEN_BUDGET_EXCEEDED, MISSING_SUBCOMMAND
• Tests: 16 new test_fread cases + tests/test_memory_ingest.sh

Changed

• Binary skip path now runs media_dispatch first; image/PDF files are read via fread-media.py instead of being skipped
• Token estimates for images now use dimension-based formula (width × height / 750) per Anthropic vision tokenization

Recent highlights

• v3.4.0 — image/PDF reading in fread, MCP media payloads, media memory ingest, review hardening, and the CAVE Starlight docs revamp
• v3.3.0 — telemetry attribution, shared run-id propagation, full-output controls, analytics rebuilds, and fedit JSON recovery hints
• v3.2.0 — fbash CLI binary added to Debian package
• v3.1.0 — fsearch SIGPIPE handling fix
• v3.0.x — fcase lifecycle overhaul, fs unified search, freplay introduction
• v2.x — fmap 50-language support, fs-engine.py refactor
• v1.x — Core 14-tool suite

Migration guides

Coming soon as the project stabilizes its public API surface.