Skip to content
OpenAgentSkill
Browsing is open · Catalog captured 2026-11-21 · Automatic updates and account services are paused.
Video & audio

create-skill-from-timeline

Reverse-engineer a finished Palmier Pro timeline into a reusable SKILL.md that recreates the same edit from the same footage with no prior context, or applies the same style to new footage with minimal tweaks. Use when the user asks to create a skill, save a template, distill this edit into a playbook, document how this video was built, or turn a timeline / chat session into a reusable workflow.

palmier-io · Static checks on file

Installation

Read the source and dependencies before running a command. Copying does not install or execute anything.

npx skills add palmier-io/palmier-skills --skill create-skill-from-timeline

About this skill


name: create-skill-from-timeline description: Reverse-engineer a finished Palmier Pro timeline into a reusable SKILL.md that recreates the same edit from the same footage with no prior context, or applies the same style to new footage with minimal tweaks. Use when the user asks to create a skill, save a template, distill this edit into a playbook, document how this video was built, or turn a timeline / chat session into a reusable workflow.

Create a skill from a timeline

A skill is a playbook another agent can follow with no prior context. Superficial notes ("add captions", "make it vertical") fail. Capture exact values and editorial judgment — what was chosen, cut, kept, and why.

Success criteria

A good skill must support both:

  1. Same footage, zero context — another agent rebuilds this video from the original media using only the skill.
  2. New footage, same style — another agent produces the same kind of edit on different sources with only light adaptation notes.

If either fails, the skill is incomplete.

Workflow checklist

- [ ] 1. Confirm scope with the user (which timeline / version)
- [ ] 2. Gather history — chat + tool trail that produced the edit
- [ ] 3. Inventory the finished timeline (get_timeline, tracks, nesting)
- [ ] 4. Inventory raw library media (get_media, inspect_media)
- [ ] 5. Diff source vs timeline — selection criteria
- [ ] 6. Read the spoken arc (get_transcript) — keep/cut rules
- [ ] 7. Capture hard numbers — layout, text, color, keyframes, cams
- [ ] 8. Infer structure + pacing + tool path
- [ ] 9. Draft SKILL.md (frontmatter + exact procedure)
- [ ] 10. Self-test: could a stranger recreate this from the skill alone?

Step 1: Confirm scope

Ask only if unclear:

  • Which timeline? (active vs a named alternate / 9:16 version)
  • Skill goal: recreate this piece, generalize a style, or both?
  • Anything off-limits (brand assets, one-off jokes, temp VO)?

Then lock that timeline with set_active_timeline if needed and re-read get_timeline.


Step 2: Recover how it was built

Prefer evidence over invention:

  1. Chat / tool history in this session — layouts chosen, caption styles, cut passes, model prompts, rejected alternatives.
  2. User intent statements — "tighter", "speaker cam only", "hook first" — these become rules, not vibes.
  3. If history is missing, reconstruct from the timeline alone and say so in the skill under Assumptions.

Record the tool path that should be reused:

Pattern in the editPrefer in the skill
Vertical / aspect reframe, PIP, split, gridapply_layout (+ anchors) — never hand set_clip_properties transforms
Multicam angle programmanage_multicam → change_cam
Dead air / fillersremove_silence then remove_words
Spoken captionsadd_captions + exact update_text style block
Authored titles / logosadd_texts with exact transform + style
Grade / FXapply_color / apply_effect with exact params
Nested versionscreate_timeline / sequence clips

Step 3: Inventory the finished piece

get_timeline()
inspect_timeline({ startFrame, endFrame, maxFrames })  // key beats

Extract and write down:

  • Canvas: fps, width, height, aspect
  • Track stack (bottom → top), names, types, what's on each
  • Clip order with [start, end), mediaRefs, speeds, trims, links
  • Nested sequences (mediaType: 'sequence') and what they contain
  • Multicam groups + angle program over time
  • Text / caption groups: content samples, captionGroupId, shared style
  • Color / effects / keyframes present on hero clips
  • Music / VO / SFX placement

inspect_timeline at the hook, mid-body, and end — verify what the viewer actually sees.


Step 4–5: Raw media vs timeline selection

get_media()
inspect_media({ mediaRef, overview: true })
inspect_media({ mediaRef, wordTimestamps: true })  // when speech drives the cut

Compare library assets to timeline clips and answer:

  • What was selected vs left unused?
  • Within a long take, which source spans were kept? (source seconds / trims)
  • Any B-roll / stills / generated assets inserted — on which spoken beats?
  • Sync relationship (multicam, dual-system, linked A/V)?

Write a Selection section: concrete rules ("keep first clean take of each claim", "drop second explanation of pricing", "B-roll on every proper noun after the hook").


Step 6: Spoken arc and cut rules

get_transcript()

Read it as prose. Document:

  • Structure — hook → body → CTA? cold open? summary then proof? straight into action?
  • Keep — sentences / claims that define the piece
  • Cut — fillers, false starts, repeated explanations, throat-clearing, long pauses
  • Join style — soft landings at sentence ends vs tight mid-thought joins
  • Silence policy — remove_silence defaults or explicit minimumPauseSeconds / speechPaddingSeconds
  • Word policy — filler list, retake handling, whether "like"/"you know" stay when intentional

After describing cuts, the remaining transcript must still read as continuous sense. Put that as an explicit verification step in the skill.


Step 7: Hard-code visual and timing details

Guessable adjectives are useless. Paste numbers.

Layout / framing

  • Exact apply_layout layout id, slot → clip mapping, fit, anchor / anchorX / anchorY
  • Track order after layout (what sits on top)
  • For alternate aspect ratios: set_project_settings / duplicate-via-create_timeline then layout — write that sequence

Text and captions

  • Font PostScript name, size, weight traits, tracking, lineSpacing, alignment, case
  • Colors (hex), outline / shadow / background blocks exactly
  • fillMode (color / footage / inverted), style.blur, widthScale / heightScale
  • Transform: alignment-relative x, y, rotation, rotationX, rotationY
  • Animation preset + highlightColor when used
  • Caption placement rel
Read the complete source ↗