Skip to content
Skills Library
Marketing & content · Planning

plan-edit

/plan-edit

Plan an edit from the analyze-project analysis report (interpretation.json and analysis-report.html) as primary evidence, obtain explicit chat approval for direction, asset planning, and execution, then produce edit.json and overlay HTML. Use for multi-asset edit planning, planning from no media through question dialogue and a provisional plan.json timeline, or choosing cuts, BGM, SFX, and B-roll from analysis.(旧名: edit-plan)

Repository
akari-video ↗
Licence
MIT
Last repository change
22.10.2026

Assessment and security

90/ 100Class: good
Format25 / 25
Frontmatter with name and description, valid name, description up to 1,024 characters, name matching the folder.
Discoverability25 / 25
Description explains when to use the skill, has a reasonable length and uses third person.
Content25 / 35
At least 150 words and at most 500 lines, headings, step lists, examples or code.
Hygiene15 / 15
No placeholders such as TODO, no Windows paths, with a licence.
Riskno signals detected in the scan
Files in this package
37
Scripts
17
HTML comments
0
URLs in SKILL.md
5
Deprecation

No signals detected. Repository changed on 22.10.2026, without references to retired models or APIs.

Licence

MIT, open: use and modify while retaining author attribution.

What to check before installing
  • 17 scripts in this package. Review them before installing, as Claude may execute them.

Scripts: bin/candidate-core.mjs, bin/canonical-json.mjs, bin/contract-semantics.mjs, bin/errors.mjs, bin/generated/contract-validators.cjs, bin/generated/runtime/equal.cjs, bin/generated/runtime/fast-deep-equal-index-9a360c74.cjs, bin/generated/runtime/ucs2length.cjs, bin/propose-cut-candidates.mjs, bin/runtime-support.mjs, bin/test/candidate-core.test.mjs, bin/test/contract-validators-regen.test.mjs, bin/

What it does and when to use it

Plan an edit from the analyze-project analysis report (interpretation.json and analysis-report.html) as primary evidence, obtain explicit chat approval for direction, asset planning, and execution, then produce edit.json and overlay HTML. Use for multi-asset edit planning, planning from no media through question dialogue and a provisional plan.json timeline, or choosing cuts, BGM, SFX, and B-roll from analysis.(旧名: edit-plan)

Original author description

Plan an edit from the analyze-project analysis report (interpretation.json and analysis-report.html) as primary evidence, obtain explicit chat approval for direction, asset planning, and execution, then produce edit.json and overlay HTML. Use for multi-asset edit planning, planning from no media through question dialogue and a provisional plan.json timeline, or choosing cuts, BGM, SFX, and B-roll from analysis.(旧名: edit-plan)

How it works

  1. Output rules for every mode
  2. Approval rules (only when autonomy is collaborative / Japanese UI: 一緒に作る)
  3. Execution order and contents
Original SKILL.md headings · 55 lines, 1,430 words
SKILL.md preview
---
name: plan-edit
description: "Plan an edit from the analyze-project analysis report (interpretation.json and analysis-report.html) as primary evidence, obtain explicit chat approval for direction, asset planning, and execution, then produce edit.json and overlay HTML. Use for multi-asset edit planning, planning from no media through question dialogue and a provisional plan.json timeline, or choosing cuts, BGM, SFX, and B-roll from analysis.(旧名: edit-plan)"
---

# Integrate editing decisions

> **Language**: Respond in the user's language — conversation, questions, approval prompts, and the reports and records you write (including decision-log rows) follow the language the user writes in.

Read lang/<content-language>.md for language-specific rules (content language: --lang → intake.json "language" → transcript language → ja; see https://github.com/AkariLabs/akari-video/blob/main/docs/contract-2026-10-11-content-language.md). If that file does not exist, read lang/en.md.
For report, decision-record, and approval-prompt writing, read the section in the user's language pack, regardless of the content language; if that pack does not exist, read lang/en.md.

> **2026-07-22 revision**: This skill no longer generates the editing decision report (fixed six-chapter HTML plus decision cards). The only formal report is the analysis report produced by [analyze-project](../analyze-project/SKILL.md); develop the direction in chat. See the opening of [report-guide.md](report-guide.md) for details and the reason for the change.

> **edit.json v2 vocabulary**: The v2 top level is exact: do not write fields other than `version` / `output` / `sources` / `tracks` / `audio` / `captions` / `thumbnail`. Do not write `beats` / `emphasis_words` / `direction` into v2 `edit.json`. Write word-level emphasis to `emphasis_words[]` at the object root of the caption SSOT, `captions.json`.

## Output rules for every mode

- Before showing the result, read [launching and checking the existing preview](../verify-edit/preview.md). Use the existing feature; do not build project-specific playback HTML, a playback UI, or audio synchronization (unless the user explicitly requests a standalone page). State the preview entry point you checked in the report.

- Base decisions on validated `analysis.json`; do not invent transcripts, frames, assets, or approvals without evidence.
- Record decisions in `decision-log.md`. For a new file, use the exact table header for the user's language; when appending to an existing file, keep its header language. Write decision and reason cells in the user's language (see [report-guide §decision_log](report-guide.md#decision_log)). Never change or delete existing rows; always append. The read-only HTML decision-record report derived from `decision-log.md` is a copy of the record without a decision UI and is no exception.
- Save still images as generated assets and include them with provenance in the report. Run i2v, avatar, and other video generation only after the corresponding still is on the timeline (for paid work, see [autonomy.md](autonomy.md) §4).
- **Save generated assets in `<project>/assets/generated/` (covered by the asset panel; revised 2026-08-12)**. Do not put them outside the project or under `<plan-dir>/`. Generate only within the project (technical enforcement is out of scope; this is a stated rule).
- Attach `<file>.meta.json` to generated assets so their origin can be reverified, recording `provenance.origin` and `provenance.generator`.
- **Place still images directly on the timeline**: put png / jpg / webp / bmp / gif images in visual-lane `items[]` as media clips (see [execution.md](execution.md), "Handling still assets"). Do not concatenate stills into a single video with ffmpeg merely to sequence them: that loses per-image editability and breaks the `edit.json` SSOT.
- Before paid or heavy generation, declare the method, reason, alternatives, and impact. Wait for an answer only for paid generation or external transmission; for free local generation, write the declaration to `decision-log.md` and proceed (in `collaborative` mode, continue to wait for an answer). Consider Codex image generation first, then Akari Cloud. Do not call OpenAI, Gemini, or similar API keys directly.
- Write **only v2 (`sources[]` + `tracks[].items[]`)** to `edit.json`. Represent lanes with `lane` and array order, and clip content with `source.kind`. Do not create the old `cuts` / `layers` / `overlays` keys. Add only fields defined by the public v2 contract.
- Use the highlight markers derived under [beats.md](beats.md) for analysis and decisions, but do not write `beats` into v2 `edit.json`.
- Derive word-level emphasis under [emphasis-detection.md](emphasis-detection.md) and write it to `emphasis_words[]` at the object root of `captions.json`. Do not write it into v2 `edit.json`.
- Use the [author-overlay](../author-overlay/SKILL.md) skill for final overlays. If it is unavailable, do not skip the rules: use the authoring rules in [CLAUDE.md](https://github.com/AkariLabs/akari-video/blob/main/CLAUDE.md) as canonical and record that choice.
- Use OpenMontage only as a reference for structural patterns. Do not copy AGPL text or code.

## Approval rules (only when autonomy is collaborative / Japanese UI: `一緒に作る`)

- **Use the analyze-project analysis report as primary evidence for editing decisions**. plan-edit itself must not generate a new decision HTML report. Decide the direction (thumbnail ideas, cut intensity, caption policy, chapter structure, etc.) in chat and record decisions in `decision-log.md`. Never change or delete existing rows; always append. The read-only HTML decision-record report derived from `decision-log.md` ([report-guide §decision_log](report-guide.md#decision_log)) is a copy without a decision UI and is no exception.
- **Stop at each direction → asset plan → execution checkpoint**. Do not treat inaction, a timeout, or blanket approval from an earlier occasion as approval now.
- After explicit approval of semantic keep/drop at Checkpoint 1, use the cut candidate bridge in [workflow.md](workflow.md) only if pause candidates are needed. Every bridge candidate is `REVIEW_REQUIRED`; it only creates a report with `approved_to_apply:false`. Do not automatically update `edit.json` or `decision-log.md`, and stop after candidate review until Checkpoint 3 execution approval. After application, do not call the work complete until a human has checked post-cut ASR, information retention, UI timing, and audio boundaries from [execution.md](execution.md) in the same version and explicitly approved `HUMAN_APPLY_GATE`.

## Execution order and contents

0. Read [autonomy.md](autonomy.md) and choose the mode from the workspace akari.md and `.akari/intake.json`. For `full-auto` ("as requested"; Japanese UI label `そのまま`) and `checkpoint` ("with suggestions"; default; Japanese UI label `提案つき`), follow autonomy.md; do not read approval steps 1–11 below. Only `collaborative` ("create together"; Japanese UI label `一緒に作る`) starts at step 1.
1. Read [workflow.md](workflow.md) to choose the analysis collection, parallel execution, and integration mode. When footage exists, read the [analyze-project](../analyze-project/SKILL.md) analysis report as primary evidence (request its creation first if absent). When there is no media, read [plan-json.md](plan-json.md) now and first complete the question dialogue → `plan.json` provisional timeline ([contract](https://github.com/AkariLabs/akari-video/blob/main/docs/contract-2026-07-20-plan-json-v0.md)).
2. Read [report-guide.md](report-guide.md) and build a recommended direction and alternatives for thumbnail ideas, cut intensity, caption policy, and chapter structure from the analysis evidence. Before presenting the direction, run [recipe.md](recipe.md)'s recall procedure on `~/.akari/recipes/` (`workflow: "edit"`) and include any recommendation with its source; it must not override this request. As part of that same step before presenting the direction, also check the `memory` declaration in `.akari/connections.json` ([contract-2026-07-25-memory-connection-v0.md](https://github.com/AkariLabs/akari-video/blob/main/docs/contract-2026-07-25-memory-connection-v0.md)). If present, start at `entry` (default `INDEX.md`), read only the `include`/`exclude` scope, and record referenced paths as provenance in `decision-log.md`. Full-text injection is forbidden. If absent, do nothing; this is not an error.
3. Present the direction in chat to the human, showing the recommendation, alternatives, reasons, and tradeoffs. Do not leave the decision at "What would you like?" Obtain explicit approval or revision instructions and append the settled decision to `decision-log.md`.
4. Read [approvals-and-generation.md](approvals-and-generation.md) and apply generation declarations, provenance, and three-stage approval. Do not execute the edit at Checkpoint 1 until explicit chat approval arrives.
5. If considering silence reduction after Checkpoint 1 approval, build the dedicated keep plan from [workflow.md](workflow.md) and present a review-only report with `bin/propose-cut-candidates.mjs`. Confirm candidate acceptance, rejection, and classification changes in chat, then append to `decision-log.md` based on human approval. Do not let the helper append records itself.

Do not read all of `edit.json` / `captions.json`; grep by id and read only matching lines ([how to read edit.json](https://github.com/AkariLabs/akari-video/blob/main/docs/guides/edit-json-access.md)). Write with a targeted edit or the edit-store script API.

6. Only after Checkpoint 3 execution approval, read [execution.md](execution.md), create v2 `edit.json` and overlay HTML, and validate them. As completion work, check the freeze procedure in [recipe.md](recipe.md) and offer to save a recipe once (offer-once) if this project has not already received that offer.
7. When deriving highlight markers, read [beats.md](beats.md) and derive candidates from `analysis.json` events / transcript. Do not write them to v2 `edit.json`; leave the mapping and evidence in the decision record.
8. When connecting approved highlights to effects, read [beat-sync.md](beat-sync.md) and build audio-lane media clips under its projection, SFX trigger, chapter-turn snap, density guardrail, and default SFX rules.
9. When choosing an expressive means for each scene, read [expression-selection.md](expression-selection.md) and follow its meaning-to-means mapping, the hard filter from the effect card's `allowed_means`, catalog search (`when_to_use` / `tags` and license check), and evidence-recording rules. This selects candidates; the asset-plan approval gate still applies.
10. When creating word-level emphasis, read [emphasis-detection.md](emphasis-detection.md), derive candidates from `analysis.json` `transcript[].words`, and write them to `emphasis_words[]` at the object root of `captions.json`. Do not write them into v2 `edit.json`; record target tier, highlight linkage, density, and evidence in the decision record.
11. When asked to insert a person or character (avatar), read [avatar-resolution.md](avatar-resolution.md) and follow staged L0/L1/L2 reading, rendition resolution, decision cards, and recording.

Read only the leaf needed for the current stage. Do not execute later stages ahead of time.
Files in this package (37)
  • SKILL.md 11 KB
  • approvals-and-generation.md 8 KB
  • autonomy.md 10 KB
  • avatar-resolution.md 15 KB
  • beat-sync.md 36 KB
  • beats.md 10 KB
  • emphasis-detection.md 18 KB
  • execution.md 15 KB
  • expression-selection.md 19 KB
  • plan-json.md 4 KB
  • recipe.md 6 KB
  • report-guide.md 22 KB
  • workflow.md 10 KB
  • bin/candidate-core.mjs 25 KB
  • bin/canonical-json.mjs 3 KB
  • bin/contract-semantics.mjs 30 KB
  • bin/errors.mjs 5 KB
  • bin/propose-cut-candidates.mjs 23 KB
  • bin/runtime-support.mjs 22 KB
  • bin/generated/contract-validators.cjs 449 KB
  • bin/generated/runtime/equal.cjs 1 KB
  • bin/generated/runtime/fast-deep-equal-index-9a360c74.cjs 1 KB
  • bin/generated/runtime/ucs2length.cjs 1 KB
  • bin/generated/runtime/licenses/ajv.txt 1 KB
  • bin/generated/runtime/licenses/fast-deep-equal.txt 1 KB
  • bin/test/candidate-core.test.mjs 16 KB
  • bin/test/contract-validators-regen.test.mjs 2 KB
  • bin/test/errors.test.mjs 3 KB
  • bin/test/l1-real-media.test.mjs 5 KB
  • bin/test/propose-cut-candidates.test.mjs 17 KB
  • bin/test/runtime-support.test.mjs 12 KB
  • bin/test/scale.test.mjs 9 KB
  • lang/en.md 3 KB
  • lang/ja.md 5 KB
  • references/asset.decisions.example.json 1 KB
  • references/cut-candidate-policy.a4-conversation-v1.json 1 KB
  • references/event.example.json 1 KB

Versions

The database has no other skills named plan-edit.

Database snapshot: 13.11.2026 · package version f97fe8e842ed (SHA-256 digest of the entire package)

Downloads in the last 30 days: 0 · ranking

Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.

Similar skills

  • 100analyze-project
    Read project assets (analysis.json) together with surrounding context (intake.json, edit.json, planning/, README, and past projects), create the interpretation layer interpretation.json, and render a read-only analysis report limited to facts and readings of the material. Use to…
  • 90video-edit-rules
    口播長影片的剪輯規範和全流程:轉寫、按腳本粗剪、B-roll 清單、風格樣片、字幕動畫、配樂音效、質檢和交付。使用者要求剪輯口播影片、粗剪、做 B-roll、或說“照規範剪”時,開剪之前先讀這份檔案。
  • 100short-form-edit
    Turn talking-head footage into a finished reel, YouTube Short, or short advertisement with curiosity-led openings, earned payoffs, story-driven cuts, transcript-synced motion graphics, moving B-roll, captions, subtle music, and synchronized sound effects. Use for short-form…
  • 100video-edit
    Plan, cut, caption, animate and render video with code. Use for any request to edit footage or make a video (short-form Reels, TikTok, Shorts, talking head, product demo, tutorial, explainer), to write an edit plan or EDL, choose cuts, B-roll or transitions, make captions or…
  • 100edit-video
    End-to-end workflow for editing a video from footage or a rough draft - look at it, transcribe it, plan the tempo, make the rough cut, add B-roll or graphics, comedic timing, sound effects and music, then export and check. Orchestrates transcribe-local, video-editing-craft,…
  • 100review-cut
    Review a human-assembled edit.json by observing only the source ranges used in its timeline, then return findings without editing. This read-only skill never modifies edit.json. Use for requests such as “How does this look?”, “Look at this cut”, or “Critique the current timeline”.(旧名: critique-cut)

Feedback and reports

Your message goes to Bartosz Sokoliński. An email address is needed only for a reply. See the privacy policy.