| Files | Last commit | Last update |
|---|---|---|
docs: state the 1.x stability guarantee and deprecation policy, pin the tool surface, verify each npm publish (#149) 1.0.0 was published by accident (see CHANGELOG 1.0.0, #145) and the number is kept. This makes the number mean what it implies: - docs/contract.md: new "Stability guarantee (1.x)" section listing what is promised through 1.x (tool ids, CLI arguments, JSON keys, exit codes, contract_version) and what is not (help text, messages, internals), and a three-step "Deprecation policy" (deprecate in a minor with a stderr warning, keep >= two minors / 90 days, remove only in the next major). - tests/test_contract.py + tests/fixtures/mcp_tools.json: the promise is enforced. Every tool's argument names and which are required are pinned; a removal, rename or newly-required argument fails CI, an addition fails until the snapshot is regenerated (UPDATE_MCP_SNAPSHOT=1) so the fixture diff shows reviewers what grew. Verified to fail on a removed argument before landing. - README: "Stability" section linking to the contract; the FFmpeg compatibility paragraph now says what CI actually runs (6.1 / 8.x / 9.x on Python 3.9) instead of implying 5.0+ is tested (#146). - release.yml: after `npm publish`, check the result rather than the exit code: the registry's `latest` must equal the version just published (with a short retry for registry lag), and an automatic bump must not have crossed a major boundary. Refs #148, #146. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs Co-authored-by: Claude <noreply@anthropic.com> | 2 months ago | |
fix(color): draw pad, text, box and background colours in BT.709; label untagged clips in mixed joins (#334) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 1 month ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
Alpha-preserving path, broadcast legal levels, smart cut (#433, carries #432 and #434) - --keep-alpha for color/graphics/overlay, background.py --transparent, alpha verification (#433) - check.py --levels, color.py --legalize (#432) - cut.py --smart, probe.py --keyframes-near (#434) | 13 days ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
Fix a contract/reality mismatch on AAC, three batch.py/render.py reliability bugs, and a non-atomic installer (#125) Continuing the same external audit report's architecture/data-integrity/ reliability findings after the security and correctness fixes in #124. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs Co-authored-by: Claude <noreply@anthropic.com> | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
FFmpeg 8 capability detection; audio extraction, audio join, sample-accurate trims, typed dynamics (0.9.1) (#20) doctor / contract: filter rows recognised by their io-spec token, so the two-character flag column of FFmpeg 8+ no longer hides every filter; unreadable listings yield `unknown`, never `missing`; fixtures captured from FFmpeg 6.1.1, 8.1.2 and 9.0.1. audio.py: audio extraction, --audio-stream, typed --compress/--limit/--gate. join.py: audio-only inputs joined as audio. cut.py: codec from the output extension, sample-accurate --accurate, precision reported. CI on pull_request with listing artifacts. Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs | 2 months ago | |
v4.0.0: the promise split by tool, platform exports at their loudness, copy joins that keep sync, and a review page (#636) * port #621 * port #627 * port #605 * fix: file names and unusable inputs say what they are (#558) A still with a % in its name that is not a pattern, a relative subtitle path with a colon, an output extension the build has no muxer for, an empty, directory, non-media or unreadable input, a closed pipe on the contract, and an input name starting with a dash. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * wip: 4.0 stability split, What 4.0 changed, changelog * wip: tool-count wording * wip: changelog after rebase * wip: pre-PR review fixes (backslash escape with a colon name, export in the verification kind, codec wording, permission denied for an unreadable file in join/audio/render, mux claim) * port #615 * port #622 * port #629 * port #626 * join --copy: predict the AAC seam delay from priming and decoded length; refuse past 50 ms in any container (#610) * review page for standalone use: plan, result and re-edit fields (#633) * port #628 * port #617 * port #624 * port #613 * export: platform presets meet the platform's loudness by default; --no-normalize keeps 3.x's report-only (#579) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fit: crop motion review asks for an excess over the cut share; no remedy that names the anchor in use (#630) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * render: an opted-out or self-levelled platform export passes --no-normalize (#579) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * join: a copy join whose sound would run ahead is refused too; the length decode holds 8 kHz bytes (#610) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * review page: check rows by status and value, an unmeasured sound says so, a cue-file field only for a cue file, no silent overwrite (#633) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * waveform: --at refuses the video-only flags it would ignore, by name Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the youtube preset now passes its whole platform check (#579) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * ci: revert proof places changed lines by the parsed test module; a timeout is not evidence (review of #635) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * v4.0.0: hand bump, the 4.0 changelog section, the ledger lines, and the one raise of the ceiling (#632) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: a normalised sound keeps its place against the picture; the loudness-miss tests keep 3.x's report-only (#579) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * budget: the ceiling counts the loudness restamp line Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the three-clip AAC copy join checks the prediction on any FFmpeg (FFmpeg 9 joins it) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the colon-name cases are POSIX names; on Windows a: is a drive Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the third clip's tone is far from the others, so its onset is its own Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * review of #636: optional probes, dash-named playlists, review-page stills and fields, dynamic-mode-only restamp, preset aliases, a removed test's decorator, doctor's exit code, docs that say what the code does Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * review page: stills live in a stills folder only this tool makes; no project input is ever in it (review of #636) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * review page: only this tool's stills are ever removed; both documents it reads are inputs; a relative output is read from the project's folder (review of #636) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * review of #636: an @ line in a test body is not a decorator; an unlistable fonts folder is unknown coverage; literal text is not a path; the page is never written over an input; docs as narrow as the code Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * review page: the page and the stills marker are written by rename, never through a link; a linked marker is refused (review of #636) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * ci: revert proof tells a deleted decorator by the base's parsed module, so an @ line inside a string is the test's (review of #636) Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: review page writes its temporary file by exclusive create, so a link at a guessable name is never written through (review of #636) The marker-symlink fixture now reaches the marker check (it has a page), and a symlink at the old temporary name is pinned. The page keeps mode 0644 rather than mkstemp's owner-only mode. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the linear-mode loudness check holds on every FFmpeg build (review of #636) FFmpeg 8.0's loudnorm closes a source gap in linear mode by itself, so the test no longer asks for the gap. It checks what this tool controls: the linear-mode encode has no restamp. Where the build keeps the gap, the test still checks that it stays at 10 s. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: review page refuses a page name hard-linked to an input, compared by file identity (review of #636) The page, each still and the stills cleanup now compare a path with the run's inputs by resolved name and by file identity (os.path.samefile). A hard link to the project JSON at index.html was replaced before; it is now refused, and the input keeps its bytes. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: revert proof places a deleted line by the base's own test map; an Igbo or Yoruba tag keeps its font; every platform preset named (review of #636) - revert_proof.py: with the base's parsed module, a deleted line belongs to its base-side test and counts while that test is in the head (a helper's decorator inside a test is the test's; a removed class is nobody's, not the test above it). - fonts.py: Vietnamese letters select a Vietnamese face unless the caller names another language (Igbo and Yoruba share these letters). - references/scripts.md: the default normalise applies to all ten platform presets, listed. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: normalising keeps a gap in the source's sound in place, on every FFmpeg build (review of #636) A gap in the sound (no samples for 2 s) was closed by the dynamic-mode restamp, and FFmpeg 8.0's loudnorm closes it in linear mode by itself: everything after the gap played that much early against the copied picture. The gap is now filled with silence before loudnorm (aresample=async=1). The test finds the silence at 8-10 s in the output in both modes. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: loudness is measured on the same gap-filled sound it normalises; the `vie` tag is Vietnamese (review of #636) - loudness.py: pass 1 measures through aresample=async=1 as pass 2 does. - fonts.py: `vie`, the three-letter form caption.py accepts, selects a Vietnamese face as `vi` does; pinned beside the Igbo and Yoruba cases. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: a short gap in the sound (from 10 ms) is filled before normalising, not only one of 100 ms or more (review of #636) swresample fills a timestamp gap only from 100 ms by default, so an 80 ms gap was closed and every later sound played 85 ms early against the copied picture (measured with the flash/beep fixture, both modes). Both loudnorm passes now run aresample=async=1:min_hard_comp=0.01. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * docs: a repository skill for building the review page (layout, numbering, composer, checklist) The editor-style review screen is planned for 4.1. The skill records what is decided: one offline HTML file that runs nothing, the 16:10 three-column layout, the numbering every element carries, options read from the contract, and how a change is checked in a real browser. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the gap check measures flash against beep, not how a build stamps its packets (review of #636) The silencedetect form failed on FFmpeg 8.0 in both modes. It checked a mechanism (silence at 8-10 s in the output) rather than what a viewer sees. The test now removes 80 ms and 0.8 s of sound from the flash/beep fixture and holds every flash to its beep after loudness.py, in both modes. On the code before the fix: 85 ms and up to 279 ms off. The comment, the changelog line and the ledger line no longer say what FFmpeg 8.0's loudnorm does: that was inferred from one failing test, not measured. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * test: the gap check asserts that its two targets ran both loudnorm modes (review of #636) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * feat: the review page is an editor-style screen that writes the instruction for the agent (#633) `npx ffmpeg-skill review project.json --result run.json` now builds one screen for a computer: sources with their probe facts, the work steps in order, a viewer that plays the output (a phone frame with the platform's safe area for a tall frame, a plain monitor otherwise), a timeline with clips, caption cues and the sound's waveform, and every measured row against what was expected. Every element has a number (S, C, T, A, M, R, P, #). Pressing one adds a row to an instruction; "Copy the instruction" gives the text to paste to the agent, with the project, the render.py command and one line per row. A work step lists every flag of its tool, read from the contract, and any of the 42 tools can be added as a step. A clip's in, out and speed are edited in place and saved with "Save project". English, Japanese, Korean and Chinese. The markup, styles and script are scripts/_review.html; _review.py measures and embeds one JSON document (no "<" of it reaches the HTML parser). The page measures nothing: a row is green only where the run's own ok is true or check.py said PASS. Shipped Python went down by 41 counted lines. Checked in Chromium at 1440x900 and 1920x1080 with a tall project with captions and an export preset, a wide one, and one with no result: no page scroll, every button named, playback, clip edits, the saved project and the copied instruction. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * chore: the code ceiling follows the review page's markup out of Python (34077 -> 34036) Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * feat: a caption's line breaks are edited in the review page and written into the instruction The caption text is a text box: Enter breaks the line, the timeline cue and the likeness under the picture follow as it is typed, and the instruction carries the break as \n with a note saying so. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix: the review page keeps its data out of the DOM and escapes every value that re-enters markup The data is the script's own value, not text read back from an element. Values that came from an attribute or a field (the selected number, a flag's key, a tool's name, the overlay colour) are escaped or accepted only when the page knows them; esc() covers both quotes; the logo is cloned from its template. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs * fix!: an English-only speech model no longer transcribes speech nobody named caption.py --transcribe and silence.py --transcribe under --engine auto, with no language named, none detectable and only an English-only Parakeet model able to run, transcribed anyway and added a note saying English was assumed: Japanese came out as English words with verified: true, and the note was read after the captions were burned. The model is now passed over before it runs and the existing english_only refusal lists it with the flag that names the language (engine_failed when a Whisper engine ran and failed). A multilingual (v3) model still runs. The note, the warning asr_language_assumed and silence.py's top-level notes are gone. A refusal where a run succeeded is a breaking change, so it goes into 4.0.0 with the others. Co-Authored-By: Claude Fable 5.2 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_013tPrsGXLuaGU7t643QNUZs --------- Co-authored-by: Claude <noreply@anthropic.com> | 15 hours ago |
ffmpeg listing fixtures
Captured or constructed stdout of ffmpeg -hide_banner -filters / -encoders / -bsfs, used by
tests/test_contract.py to prove that capability detection (doctor) reads every layout FFmpeg
has printed. CI uploads each runner's listings and doctor --json as the ffmpeg-listings-<os>
artifact; copy a new layout from there into this directory and add it to the test table.
| File | Origin |
|---|---|
ffmpeg_filters_6.1.txt, ffmpeg_encoders_6.1.txt, ffmpeg_bsfs_6.1.txt |
captured from FFmpeg 6.1.1 (Ubuntu 24.04 apt; same output as the ubuntu-latest CI runner) |
ffmpeg_filters_8.1.2_macos.txt, ffmpeg_encoders_8.1.2_macos.txt, ffmpeg_bsfs_8.1.2_macos.txt |
captured from FFmpeg 8.1.2 (Homebrew bottle, macos-latest CI runner). This build has no libfreetype / libass / libzimg, so drawtext, subtitles, ass and zscale are absent from the listing; the tests expect doctor to report exactly those as missing |
ffmpeg_filters_9.0.1_windows.txt, ffmpeg_encoders_9.0.1_windows.txt, ffmpeg_bsfs_9.0.1_windows.txt |
captured from FFmpeg 9.0.1 (gyan.dev essentials build, windows-latest CI runner, CRLF line endings kept as captured) |
ffmpeg_filters_7.1_constructed.txt |
constructed: the 6.1 capture unchanged, kept under its own name so the version matrix names FFmpeg 7 explicitly (7.x prints the 6.x layout: three flag characters T.. .S. ..C). Not a capture of an FFmpeg 7 binary. |
ffmpeg_filters_garbage.txt |
not a listing at all: must yield unknown, never missing |
What the FFmpeg 8.1.2 and 9.0.1 captures show (an earlier constructed 8.0 file had guessed a
different legend and was replaced by the capture): the legend still prints three-character keys
(T.. = Timeline support, .S. = Slice threading) while the rows carry two ( TS aap AA->A),
and a ------ separator line follows the legend. The parser keys on the io-spec token (A->A,
AA->A, |->V, N->N) and ignores the flag column and the legend.
The 7.1 file is the one remaining construction: it reproduces the 6.x row format with the 6.1 filter set and is not a capture of an FFmpeg 7 binary. Replace it with a capture when one is at hand and keep the name so the tests keep working.
mcp_tools.json
Not an ffmpeg listing: the frozen 1.x MCP/CLI surface (per tool, the argument names and which are required), read by tests/test_contract.py's test_mcp_tool_surface_matches_the_frozen_1x_snapshot. Regenerate with UPDATE_MCP_SNAPSHOT=1 python3 tests/test_contract.py only for additions; see docs/contract.md, "Stability guarantee (1.x)".