kajisho5/ffmpeg-skill v2.21.0 on GitHub

Automated release: version and notes generated from pull requests merged since 2.20.0.

One table of verdict thresholds. references/scripts.md § "Verdict thresholds" lists the numbers behind verified and check.py's rows: true-peak and loudness tolerances, the length tolerances of loudness.py and join.py, cut.py's keyframe tolerance, the silence, digital-silence and dropout levels, --content's shares and filler timings. It also gives each platform's loudness target, tolerance, true-peak ceiling and maximum length. Before, these numbers were split across the code, docs/contract.md and docs/design-decisions.md, and some were in no doc at all. Each row names the constant or flag it comes from, and a test holds the table to the code.
README FAQ: "Why 42 tools?" It explains why there is one tool per job, and that over MCP an agent sees the core 12 by default and calls the other 30 by name. A table splits the 42 by the Tools section's groups and marks the 12, and a test holds it to that section and to MCP_CORE_TOOLS.

The -50 dBFS silent-track level is one constant. join.py, audio.py and waveform.py took their --silence-threshold default from a literal; it is now _common.probe.SILENT_PEAK_DBFS, the level check.py already used. loudness.py's ±1.0 LU and join.py's duration tolerance are named constants too. No value changed.
README FAQ "Why 42 tools?", and no release number in the thresholds heading (#374)
One table of verdict thresholds, held to the code (#373)

What's Changed

One table of verdict thresholds, held to the code by @kajisho5 in #373
README FAQ "Why 42 tools?", and no release number in the thresholds heading by @kajisho5 in #374

Full Changelog: v2.20.0...v2.21.0