{
  "schemaVersion": "1",
  "revision": "22ce5d34dbcb5a78",
  "id": "codex-video-workflows-reference",
  "slug": "video-workflows-reference",
  "title": "Video analysis and evidence workflows: practical reference",
  "summary": "Repeatable video-to-prompt workflow covering source preservation, hashing and media receipts, chunking, Oracle/Gemini analysis, timestamped evidence lanes, sentinel checks, keyframes, omissions review, privacy gates, and downstream implementation, QA, and…",
  "category": "Context",
  "tags": [
    "video",
    "oracle",
    "evidence"
  ],
  "sources": [
    {
      "label": "Portable Codex workflow resources",
      "url": "https://lapis-fossil-vsbc.here.now/downloads/codex-power-user-workshop-2026-06-27.bundle"
    }
  ],
  "sourceUnits": [
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Agent Setup Prompt",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Big Idea",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Try This Now",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Tool Reference",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/When To Reach For This",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/How This Used To Work",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Core Pattern",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Short Path vs Long Path",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Run Folder Shape",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Preserve The Source",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Evidence Lanes",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Prompt With A Sentinel",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Analyze With Gemini",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Verify The Analysis",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Omissions Pass",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Turn Video Into Prompt Material",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/Ask The Agent This",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/What To Confirm / How To Know It Worked",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/No Source Hash",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/No `ffprobe` Output",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/Oracle Call Used API Mode",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/File Path Was Relative",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/Sentinel Missing",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/Analysis Only Summarizes Transcript",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/Visual Claim Has No Keyframe",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/URL, Tool, Command, Or File Appears Without Evidence",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/If It Goes Wrong/Long Recording Was Not Chunked",
    "downloads/codex-power-user-workshop-2026-06-27.bundle!/cheatsheets/07-video-as-prompts.md#07. Video As Prompts/References And Resources",
    "downloads/codex-power-user-workshop-2026-06-27.bundle"
  ],
  "collection": "codex",
  "categoryId": "context",
  "wordCount": 2668,
  "url": "/topics/codex-video-workflows-reference/",
  "markdownUrl": "/markdown/codex-video-workflows-reference.md?raw=1&v=22ce5d34dbcb5a78",
  "jsonUrl": "/api/v1/documents/codex-video-workflows-reference.json",
  "headings": [
    {
      "id": "agent-setup-prompt",
      "text": "Agent Setup Prompt",
      "level": 2
    },
    {
      "id": "big-idea",
      "text": "Big Idea",
      "level": 2
    },
    {
      "id": "try-this-now",
      "text": "Try This Now",
      "level": 2
    },
    {
      "id": "tool-reference",
      "text": "Tool Reference",
      "level": 2
    },
    {
      "id": "when-to-reach-for-this",
      "text": "When To Reach For This",
      "level": 2
    },
    {
      "id": "how-this-used-to-work",
      "text": "How This Used To Work",
      "level": 2
    },
    {
      "id": "core-pattern",
      "text": "Core Pattern",
      "level": 2
    },
    {
      "id": "short-path-vs-long-path",
      "text": "Short Path vs Long Path",
      "level": 2
    },
    {
      "id": "run-folder-shape",
      "text": "Run Folder Shape",
      "level": 2
    },
    {
      "id": "preserve-the-source",
      "text": "Preserve The Source",
      "level": 2
    },
    {
      "id": "evidence-lanes",
      "text": "Evidence Lanes",
      "level": 2
    },
    {
      "id": "prompt-with-a-sentinel",
      "text": "Prompt With A Sentinel",
      "level": 2
    },
    {
      "id": "analyze-with-gemini",
      "text": "Analyze With Gemini",
      "level": 2
    },
    {
      "id": "verify-the-analysis",
      "text": "Verify The Analysis",
      "level": 2
    },
    {
      "id": "omissions-pass",
      "text": "Omissions Pass",
      "level": 2
    },
    {
      "id": "turn-video-into-prompt-material",
      "text": "Turn Video Into Prompt Material",
      "level": 2
    },
    {
      "id": "ask-the-agent-this",
      "text": "Ask The Agent This",
      "level": 2
    },
    {
      "id": "what-to-confirm-how-to-know-it-worked",
      "text": "What To Confirm / How To Know It Worked",
      "level": 2
    },
    {
      "id": "if-it-goes-wrong",
      "text": "If It Goes Wrong",
      "level": 2
    },
    {
      "id": "no-source-hash",
      "text": "No Source Hash",
      "level": 3
    },
    {
      "id": "no-ffprobe-output",
      "text": "No `ffprobe` Output",
      "level": 3
    },
    {
      "id": "oracle-call-used-api-mode",
      "text": "Oracle Call Used API Mode",
      "level": 3
    },
    {
      "id": "file-path-was-relative",
      "text": "File Path Was Relative",
      "level": 3
    },
    {
      "id": "sentinel-missing",
      "text": "Sentinel Missing",
      "level": 3
    },
    {
      "id": "analysis-only-summarizes-transcript",
      "text": "Analysis Only Summarizes Transcript",
      "level": 3
    },
    {
      "id": "visual-claim-has-no-keyframe",
      "text": "Visual Claim Has No Keyframe",
      "level": 3
    },
    {
      "id": "url-tool-command-or-file-appears-without-evidence",
      "text": "URL, Tool, Command, Or File Appears Without Evidence",
      "level": 3
    },
    {
      "id": "long-recording-was-not-chunked",
      "text": "Long Recording Was Not Chunked",
      "level": 3
    },
    {
      "id": "references-and-resources",
      "text": "References And Resources",
      "level": 2
    },
    {
      "id": "related-resources",
      "text": "Related resources",
      "level": 2
    }
  ],
  "markdown": "# Video analysis and evidence workflows: practical reference\n\n## Agent Setup Prompt\n\n```text\nSet up a repeatable video-as-prompt run for this project before doing any full video analysis.\n\n1. Resolve the `oracle-video` skill first. Check for `skills/oracle-video/SKILL.md`; if it is not bundled, check the common local install path at `.agents/skills/oracle-video/SKILL.md`. Read the skill and any directly referenced repeatable video process file before creating outputs. If the skill is missing, stop and report the missing path checks.\n\n2. Create a run id like `<YYYY-MM-DD>-<short-slug>` and a run folder at `runs/<run-id>/` with this structure: `inputs/`, `receipts/`, `chunks/01/`, `analysis/`, `final/`, `keyframes/`, and `contact-sheets/`.\n\n3. Accept either a local video path or a video URL. For a local file, copy it to `inputs/source-video.<ext>` and keep that copy immutable. For a URL, first save metadata with `yt-dlp --dump-single-json --skip-download --no-check-formats` to `receipts/source.metadata.json`, then download with `yt-dlp -f \"bv*+ba/best\"` into `inputs/`.\n\n4. Write a source hash with `shasum -a 256` to `inputs/source-video.sha256`. Confirm the file exists and is non-empty.\n\n5. Capture media facts with `ffprobe -hide_banner -show_format -show_streams -of json` and save them to `receipts/ffprobe.source.json`. Confirm the file exists and is non-empty. If `ffprobe`, `ffmpeg`, or `yt-dlp` is missing, stop setup and report the missing tool plus the failed command.\n\n6. Prepare `chunks/01/chunk.mp4`. For a short video, copy the source video into that path. For a long video, create additional numbered chunk folders as needed and save a timestamp mapping note before any analysis. Use absolute paths for every video file that will be attached to Oracle.\n\n7. Write `chunks/01/gemini-prompt.md`. It must request separate lanes for transcript evidence, chat evidence, visual evidence, and inference. It must require preserved timestamps, forbid invented URLs, tools, commands, files, UI labels, and speakers, and end with the exact final line `END_OF_VIDEO_ANALYSIS`.\n\n8. Write `chunks/01/expected-sentinel.txt` containing only `END_OF_VIDEO_ANALYSIS`. Confirm the sentinel appears in `gemini-prompt.md` and that the expected sentinel file is non-empty.\n\n9. Prepare `chunks/01/oracle-call.json` or equivalent setup notes with the required Gemini browser-mode contract: `engine: \"browser\"`, `model: \"gemini-3-pro\"`, `browserAttachments: \"always\"`, `files` containing an absolute path to `chunks/01/chunk.mp4`, and a stable slug. Do not use API mode for video analysis.\n\n10. Generate keyframes and a contact sheet if `ffmpeg` is available. Save frames under `keyframes/` using a low-frequency extraction such as `fps=1/10`, then save a tiled contact sheet under `contact-sheets/`. If generation is skipped, save the exact commands to run later and report why they were not run.\n\n11. Create evidence lane files under `analysis/`: `transcript.md`, `chat.md`, `visual-evidence.md`, and `inference.md`. Keep spoken claims, chat or comments, visible facts, and reasoned conclusions separate. Label inference clearly until code, runtime behavior, or visual checks support it.\n\n12. Create `analysis/omissions-pass-prompt.md`. It should ask only for missing visible UI states, skipped timestamps, unsupported claims, invented details, uncertainty, and patch-style corrections. It must require the response to end with `END_OF_VIDEO_ANALYSIS`.\n\n13. Create downstream artifact stubs under `final/`, such as `implementation-prompt.md`, `qa-checklist.md`, and `variation-matrix.md`. Leave evidence-dependent fields blank until verified analysis exists.\n\n14. Enforce privacy boundaries before any upload or external analysis. Do not print full environment output, tokens, cookies, API keys, private endpoints, resolved secret values, or raw private config. Do not copy secret helper files into the run folder. If the video, metadata, transcript, or chat text contains credentials or private config, stop and report the issue instead of uploading.\n\n15. Run a setup verification gate before Gemini analysis. Report `OK` or `MISSING` for: `oracle-video` skill resolved, source video exists, source hash non-empty, `ffprobe.source.json` non-empty, URL metadata present when the input is a URL, chunk file exists, Gemini prompt exists, sentinel file exists, Oracle call uses browser mode with `gemini-3-pro` and `browserAttachments: \"always\"`, Oracle file path is absolute, evidence lane files exist, keyframes/contact sheet exist or commands are saved, omissions prompt exists, and downstream stubs exist.\n\n16. Stop before full video analysis unless every required setup check passes. Return the run id, absolute run folder path, commands run, verification report, failed checks, and the smallest next fix if anything is missing. Only say it is ready for Gemini analysis after the setup gate passes.\n```\n\n## Big Idea\n\nVideo is prompt material.\n\n## Try This Now\n\nFor a short bug repro video, create a run folder and save a source hash.\nWrite the Gemini prompt with separate lanes: transcript, chat, visual evidence, inference.\nEnd the prompt with `END_OF_VIDEO_ANALYSIS` before analysis starts.\n\nA screen recording can show things a written prompt usually loses:\n\n- order of operations\n- cursor movement\n- hesitation\n- timing\n- hidden assumptions\n- actual UI state\n- terminal output\n- visible browser pages\n- misclicks\n- loading states\n- visual hierarchy\n- broken affordances\n- spoken explanation\n- chat context\n\nA transcript can tell you what someone said. A video can show what happened.\n\n## Tool Reference\n\nUse `oracle-video` when the task depends on what is visible or audible in a recording.\n\nConceptually:\n\n```text\noracle-packx bundles code/text context for ChatGPT.\noracle-video attaches video and sends it to Gemini through Oracle browser mode.\n```\n\nCorrect Oracle call shape for video analysis:\n\n```js\nmcp__oracle__consult({\n  prompt: \"[video-analysis-slug]\\n\\n{contents of chunks/01/gemini-prompt.md}\",\n  files: [\"/absolute/path/to/runs/bug-repro/chunks/01/chunk.mp4\"],\n  engine: \"browser\",\n  model: \"gemini-3-pro\",\n  browserAttachments: \"always\",\n  slug: \"video-analysis-slug\"\n})\n```\n\nRequired:\n\n```text\nengine: \"browser\"\nmodel: \"gemini-3-pro\"\nbrowserAttachments: \"always\"\nabsolute file path\n```\n\nDo not use API mode for video analysis.\n\n## When To Reach For This\n\nUse video when the important context is visible, temporal, or hard to describe.\n\n| Use Video For | Why |\n| --- | --- |\n| bug reproduction clip | captures exact UI state, clicks, errors, timing |\n| design feedback video | preserves visual hierarchy and spoken intent |\n| app walkthrough | turns observed workflow into requirements and QA checks |\n| loading/timing bug | shows delay, spinner, race, or order of events |\n| terminal/browser workflow | preserves commands, pages, output, and sequence |\n| long tutorial or workshop recording | can become chapters, exercises, references, and follow-up prompts |\n\nSkip video when:\n\n- a short log already proves the issue\n- a failing test names the exact bug\n- the needed context is one code path\n- the video source cannot be preserved\n- claims cannot be verified against timestamps, keyframes, or source output\n\n## How This Used To Work\n\nWhat you used to expect:\n\n```text\nWatch the video, summarize from memory, then ask the agent to fix the vague thing.\n```\n\nWhat the workflow gives you now:\n\n```text\nPreserve the source, capture hashes and media facts, analyze with Gemini, keep\nevidence lanes separate, verify claims, then turn the result into a grounded\nprompt.\n```\n\nWhat you used to expect:\n\n```text\nA transcript is the prompt.\n```\n\nWhat the workflow gives you now:\n\n```text\nThe transcript is one evidence lane. Visual state, timing, chat, cursor motion,\nand inference stay separate until verified.\n```\n\n## Core Pattern\n\n```text\nsource media\n-> source receipts\n-> Gemini analysis\n-> evidence lanes\n-> verification\n-> downstream prompt\n```\n\nDo not skip the middle.\n\n```text\nvideo without source receipts = hard to trust\nanalysis without timestamps = hard to verify\ntranscript without visuals = incomplete\nclaims without keyframes = risky\nprompt without verification = hallucination bait\n```\n\n## Short Path vs Long Path\n\n| Path | Use For | Shape |\n| --- | --- | --- |\n| short path | bug repros, design feedback clips, quick app walkthroughs | preserve source, one Gemini pass, keyframes, downstream artifact |\n| long path | long tutorials, workshop recordings, multi-section demos | source receipts, segment/chunk, analyze chunks, omissions pass, reconcile, final packet |\n\nShort path output examples:\n\n- repro steps\n- implementation prompt\n- design-change brief\n- QA checklist\n- requirements note\n\nLong path output examples:\n\n- chapter index\n- lesson notes\n- exercise packet\n- resource list\n- follow-up prompts\n- Section 6 variation matrix\n\n## Run Folder Shape\n\nUse a run folder so the video stays tied to its analysis.\n\n```text\nruns/<date>-<slug>/\n  inputs/\n    source-video.mp4\n    source-video.sha256\n  receipts/\n    ffprobe.source.json\n  chunks/\n    01/\n      chunk.mp4\n      gemini-prompt.md\n      expected-sentinel.txt\n      output.log\n      analysis.md\n  analysis/\n  final/\n  keyframes/\n  contact-sheets/\n```\n\nCreate the folders:\n\n```bash\nRUN_ID=\"checkout-repro\"\nRUN_DIR=\"runs/$RUN_ID\"\nmkdir -p \"$RUN_DIR\"/{inputs,receipts,analysis,final,keyframes,contact-sheets}\nmkdir -p \"$RUN_DIR/chunks/01\"\n```\n\n## Preserve The Source\n\nKeep the original media immutable.\n\nFor a local file:\n\n```bash\ncp /path/to/source-video.mp4 \"$RUN_DIR/inputs/source-video.mp4\"\nshasum -a 256 \"$RUN_DIR/inputs/source-video.mp4\" \\\n  > \"$RUN_DIR/inputs/source-video.sha256\"\n```\n\nFor a URL source, capture metadata first when useful:\n\n```bash\nyt-dlp --dump-single-json --skip-download --no-check-formats \"$URL\" \\\n  > \"$RUN_DIR/receipts/source.metadata.json\"\n```\n\nThen download:\n\n```bash\nyt-dlp -f \"bv*+ba/best\" -o \"$RUN_DIR/inputs/source-video.%(ext)s\" \"$URL\"\n```\n\nCapture media facts:\n\n```bash\nffprobe -hide_banner -show_format -show_streams -of json \\\n  \"$RUN_DIR/inputs/source-video.mp4\" \\\n  > \"$RUN_DIR/receipts/ffprobe.source.json\"\n```\n\nConfirm:\n\n```bash\ntest -s \"$RUN_DIR/inputs/source-video.sha256\" && echo \"sha256:OK\"\ntest -s \"$RUN_DIR/receipts/ffprobe.source.json\" && echo \"ffprobe:OK\"\n```\n\n## Evidence Lanes\n\nKeep these lanes separate:\n\n| Lane | What It Means | Example |\n| --- | --- | --- |\n| transcript | spoken words | “The save button disappears.” |\n| chat | chat messages or comments | “This only happens after refresh.” |\n| visual evidence | visible UI state | save button hidden at `00:01:34` |\n| inference | reasoned conclusion | likely state reset after token refresh |\n\nPlain rule:\n\n```text\nA spoken claim is not a visible fact.\nA visible fact is not automatically the cause.\nAn inference must stay labeled until code or runtime proof confirms it.\n```\n\n## Prompt With A Sentinel\n\nSave the Gemini prompt beside the chunk.\n\n```bash\ncat > \"$RUN_DIR/chunks/01/gemini-prompt.md\" <<'EOF'\nAnalyze this video as prompt material for a software project.\n\nReturn:\n\n1. short summary\n2. timestamped workflow\n3. transcript evidence\n4. chat evidence\n5. visual evidence\n6. inferences, clearly labeled\n7. likely downstream artifacts\n8. omissions or uncertainty\n\nRules:\n\n- Do not invent URLs, tools, commands, files, UI labels, or speakers.\n- Preserve original timestamps.\n- Separate visible facts from spoken claims and inferences.\n- Include the exact sentinel below as the last line.\n\nEND_OF_VIDEO_ANALYSIS\nEOF\n\nprintf 'END_OF_VIDEO_ANALYSIS\\n' \\\n  > \"$RUN_DIR/chunks/01/expected-sentinel.txt\"\n```\n\nConfirm:\n\n```bash\nrg -n \"END_OF_VIDEO_ANALYSIS\" \"$RUN_DIR/chunks/01/gemini-prompt.md\"\ntest -s \"$RUN_DIR/chunks/01/expected-sentinel.txt\" && echo \"sentinel:OK\"\n```\n\n## Analyze With Gemini\n\nUse Oracle browser mode and attach the video.\n\n```js\nmcp__oracle__consult({\n  prompt: \"[checkout-repro-video]\\n\\n{contents of runs/.../chunks/01/gemini-prompt.md}\",\n  files: [\"/absolute/path/to/runs/checkout-repro/chunks/01/chunk.mp4\"],\n  engine: \"browser\",\n  model: \"gemini-3-pro\",\n  browserAttachments: \"always\",\n  slug: \"checkout-repro-video\"\n})\n```\n\nAfter the run, read the local output:\n\n```bash\nsed -n '1,220p' ~/.oracle/sessions/checkout-repro-video/output.log\nrg -n \"END_OF_VIDEO_ANALYSIS\" ~/.oracle/sessions/checkout-repro-video/output.log\n```\n\nIf the output is clipped in a tool response, treat `output.log` as the local source to inspect.\n\n## Verify The Analysis\n\nCheck that the answer is grounded.\n\n| Check | Command Or Evidence |\n| --- | --- |\n| source hash exists | `test -s \"$RUN_DIR/inputs/source-video.sha256\"` |\n| media facts exist | `test -s \"$RUN_DIR/receipts/ffprobe.source.json\"` |\n| prompt saved | `test -s \"$RUN_DIR/chunks/01/gemini-prompt.md\"` |\n| sentinel present | `rg -n \"END_OF_VIDEO_ANALYSIS\" output.log` |\n| timestamps preserved | compare analysis timestamps to source duration |\n| visual claims checked | keyframes or contact sheet |\n| invented facts removed | no unsupported URLs/tools/commands/files |\n| omissions captured | missing sections are listed as uncertainty |\n\nExtract keyframes for visual checks:\n\n```bash\nmkdir -p \"$RUN_DIR/keyframes\"\nffmpeg -i \"$RUN_DIR/inputs/source-video.mp4\" \\\n  -vf \"fps=1/10\" \\\n  \"$RUN_DIR/keyframes/frame-%04d.jpg\"\n```\n\nCreate a contact sheet:\n\n```bash\nffmpeg -pattern_type glob -i \"$RUN_DIR/keyframes/*.jpg\" \\\n  -vf \"scale=320:-1,tile=4x4\" \\\n  \"$RUN_DIR/contact-sheets/source-contact-sheet.jpg\"\n```\n\n## Omissions Pass\n\nRun an omissions pass when coverage is incomplete.\n\nAsk:\n\n```text\nReview the prior video analysis for omissions.\n\nFocus only on missing visible UI states, skipped timestamps, unsupported claims,\ninvented details, and uncertainty. Do not rewrite the whole analysis. Return a\npatch-style list of corrections and end with END_OF_VIDEO_ANALYSIS.\n```\n\nUse the omissions pass when:\n\n- the first analysis only summarizes the transcript\n- key UI states are missing\n- timestamps are vague\n- the model invents tools, URLs, files, or labels\n- a long recording was not chunked\n- the downstream prompt depends on visual details\n\n## Turn Video Into Prompt Material\n\nCommon downstream artifacts:\n\n| Video Source | Downstream Artifact |\n| --- | --- |\n| bug reproduction | repro steps, expected vs actual behavior, implementation prompt |\n| design feedback | visual acceptance criteria, change brief, QA checklist |\n| app walkthrough | requirements, acceptance criteria, open questions |\n| terminal workflow | command sequence, failure point, debugging prompt |\n| long recording | chapters, exercises, resources, follow-up packet |\n| UI alternatives | Section 6 variation matrix |\n\nBug prompt shape:\n\n```markdown\n# Implementation Prompt From Video\n\n## Source\n\n- video:\n- source hash:\n- ffprobe:\n- analysis:\n- keyframes:\n\n## Observed Failure\n\n- timestamp:\n- visual evidence:\n- transcript/chat evidence:\n- expected behavior:\n- actual behavior:\n\n## Request\n\nImplement the smallest safe fix.\n\n## Constraints\n\n- do not change:\n- preserve:\n\n## Verification\n\n- command:\n- what passing means:\n```\n\nVariation seed shape:\n\n```markdown\n# Video-Derived Variation Matrix\n\n## Goal\n\n## Evidence From Video\n\n- timestamp:\n- observation:\n- keyframe:\n- why it matters:\n\n## Invariants\n\n- same source evidence:\n- same target flow:\n- same verification command:\n- same review criteria:\n\n## Variable Under Test\n\n- variable:\n- why this variable matters:\n- what is out of scope:\n```\n\n## Ask The Agent This\n\nAnalyze a bug reproduction:\n\n```text\nUse the video analysis workflow.\n\nPreserve the source video, save a source hash and ffprobe output, write a Gemini\nprompt with END_OF_VIDEO_ANALYSIS, analyze the video through Oracle browser mode\nwith Gemini, then separate transcript evidence, visual evidence, chat evidence,\nand inference. Turn the verified result into repro steps and an implementation\nprompt for the smallest safe fix.\n```\n\nReview design feedback:\n\n```text\nTurn this design feedback video into a change brief.\n\nKeep visible UI facts separate from spoken opinions. Save keyframes for the\nvisual claims. Produce acceptance criteria, a QA checklist, and open questions.\nDo not invent UI labels or routes that are not visible or spoken.\n```\n\nCreate a Section 6 seed:\n\n```text\nUse this verified video analysis to create a controlled variation matrix.\n\nChoose one variable under test, preserve the same source evidence and\nverification command for every candidate, and write three candidate settings.\n```\n\n## What To Confirm / How To Know It Worked\n\n| Goal | Confirmation |\n| --- | --- |\n| source preserved | original media exists under `inputs/` |\n| source hash saved | `source-video.sha256` exists and is non-empty |\n| media facts saved | `ffprobe.source.json` exists |\n| prompt saved | `chunks/01/gemini-prompt.md` exists |\n| sentinel required | `expected-sentinel.txt` exists |\n| Oracle call shape correct | `engine: \"browser\"`, `model: \"gemini-3-pro\"`, `browserAttachments: \"always\"` |\n| output inspected | `~/.oracle/sessions/<slug>/output.log` read |\n| sentinel present | `END_OF_VIDEO_ANALYSIS` found in output |\n| lanes separated | transcript, chat, visual evidence, inference are separate |\n| visual claims checked | keyframes/contact sheet exists |\n| unsupported claims removed | no invented URLs/tools/commands/files |\n| downstream artifact saved | repro, requirements, QA, brief, lesson, or matrix exists |\n\n## If It Goes Wrong\n\n### No Source Hash\n\nCreate it before analysis:\n\n```bash\nshasum -a 256 \"$RUN_DIR/inputs/source-video.mp4\" \\\n  > \"$RUN_DIR/inputs/source-video.sha256\"\n```\n\n### No `ffprobe` Output\n\nRun:\n\n```bash\nffprobe -hide_banner -show_format -show_streams -of json \\\n  \"$RUN_DIR/inputs/source-video.mp4\" \\\n  > \"$RUN_DIR/receipts/ffprobe.source.json\"\n```\n\n### Oracle Call Used API Mode\n\nRerun with browser mode.\n\n```text\nengine: \"browser\"\nmodel: \"gemini-3-pro\"\nbrowserAttachments: \"always\"\n```\n\n### File Path Was Relative\n\nUse an absolute video path in `files`.\n\n```bash\nrealpath \"$RUN_DIR/chunks/01/chunk.mp4\"\n```\n\n### Sentinel Missing\n\nRead `output.log`. If the sentinel is truly missing, rerun or ask for an omissions pass.\n\n```bash\nrg -n \"END_OF_VIDEO_ANALYSIS\" ~/.oracle/sessions/<slug>/output.log\n```\n\n### Analysis Only Summarizes Transcript\n\nRun an omissions pass focused on visual evidence:\n\n```text\nFind missing visual UI states, cursor actions, timing, browser/terminal state,\nand unsupported claims. Preserve timestamps.\n```\n\n### Visual Claim Has No Keyframe\n\nExtract keyframes or create a contact sheet. Downgrade unsupported claims to inference.\n\n### URL, Tool, Command, Or File Appears Without Evidence\n\nRemove it or label it as inference. Do not let invented details become prompt constraints.\n\n### Long Recording Was Not Chunked\n\nCreate chunks and analyze them separately. Preserve original timestamps across chunks.\n\n## References And Resources\n\n- `oracle-video` local skill: `skills/oracle-video/SKILL.md` if bundled, or common local skill install under `.agents/skills/oracle-video/SKILL.md`\n- `yt-dlp`: https://github.com/yt-dlp/yt-dlp\n- FFmpeg documentation: https://ffmpeg.org/documentation.html\n- `ffprobe` documentation: https://ffmpeg.org/ffprobe.html\n- Gemini app: https://gemini.google.com/app\n\n## Related resources\n\n- [Companion reference](https://lapis-fossil-vsbc.here.now/downloads/codex-power-user-workshop-2026-06-27.bundle)\n- [Resource directory and portable download](/topics/codex-resources/)"
}
