agentleFS
tamtree-aitamtree-ai/stick-stage/skills/crew-episode/SKILL.md

crew-episode

Skill2 starsChanged 15 days agoMITView on GitHub

Write, stage and render an episode of the relatable three-person channel (working name "Same Three People"): Mitch, Ollie and Maya on the commuter train, in the office meeting room or at a backyard BBQ have a 20-28 s everyday conversation with a 2-second hook, a laugh at the two-thirds mark and Maya's quiet closing thought as the train stops. Covers the premise test, beat map, line rules, shots, the platform cut, running gags, packaging (POV card, title, caption), multi-part rules, render and hand-back. Use when asked for a crew / Same Three People / train-channel episode, the next episode of this channel, or to restage crew-we-launch-monday or crew-millionaire-at-94. The team writes or approves the final jokes; drafts are placeholders.

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 6 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈6

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read first: src/data/series/crew.json (the series bible), the cast bibles src/data/characters/crew-mitch.md, crew-ollie.md, crew-maya.md, …

    line 12
  2. Very specific:

    Isolation: Use only crew-* sets, characters and parts, the crew series, and per-skit settings.

    line 23
  3. Distinctive:

    No moral, no "and that's why…", no advice.

    line 49
  4. Distinctive:

    If 3 is a lecture or 4 is nobody, change the premise.

    line 52
  5. Very specific:

    No narrator, no "fun fact", no moral.

    line 79
  6. Distinctive:

    Use at most one per episode and never as the main joke.

    line 131
  7. Distinctive:

    Default: every episode is a complete story.

    line 145
  8. Distinctive:

    Never re-use footage, lines or a recap of the earlier part (a one-line callback is fine).

    line 150
  9. Distinctive:

    Never more than 3 parts. Never end on "follow for part 2": the episode ends on the thought.

    line 152
  10. Very specific:

    Check the 2-second catch from voice.json: leadInMs + line 1 durationMs must be under ~2,400 ms with say (under 2,000 ms with final voices).

    line 165

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Crew episode (Same Three People)

Three friends who commute together talk about something everyone does. It is funny, and the last line leaves the viewer thinking. The audience test is a share: "this is literally us", "tag your Mitch". If a viewer can't picture a friend who does this, the premise is wrong.

Read first: src/data/series/crew.json (the series bible), the cast bibles src/data/characters/crew-mitch.md, crew-ollie.md, crew-maya.md, the episode log ../planning/2026-11-10-relatable-four-channel/episodes/README.md (what's been used: topics, fault, laugh lines, running-gag state), and the reference episodes public/skits/crew-we-launch-monday/ and public/skits/crew-millionaire-at-94/. General staging: skills/skit-director/SKILL.md; skit format: docs/skits.md.

Hard rules

  1. The team writes or approves the jokes. Draft every line so the episode renders, but say in the hand-back which lines are placeholders. Never present a draft as final.
  2. Isolation. Use only crew-* sets, characters and parts, the crew series, and per-skit settings. Never edit an engine default, a shared style or another channel's data to fix this channel. A new place is a new crew-* set.
  3. Each episode stands alone. A viewer who has seen nothing else must get every line. No "as we said last week", no unexplained in-jokes; a running gag must also work as a plain joke.
  4. Out of bounds: real people, real brands as the butt of the joke, men-vs-women framing, money, medical or legal advice. Petty is the point; mean is not.

The cast (never swap roles)

IdWhoJob in every episode
mitch (crew-mitch, stands far-left, lean-in)Australian, the Chaos OneRelaxed, reckless, never worried. Says the bad idea as if it's obvious.
ollie (crew-ollie, sits left)English, the OverthinkerPanics, plans, apologises, over-prepares.
maya (crew-maya, sits right)The DeadpanFew words. Undercuts both. Always has the last line.

Mitch and Ollie are the double act (one doesn't care, one cares too much). Who's at fault rotates: the log records it; don't make the same one the butt two episodes running. Maya is sometimes wrong too, but she still lands the thought.

Premise test (write these down before any line)

  1. The behaviour, in under 8 words, that most viewers do or know someone who does ("never checks the bank balance").
  2. The two sides: what Mitch does and what Ollie does about it. Opposites are the engine.
  3. The thought: the one sentence Maya leaves the viewer with. It reframes, it doesn't preach. No moral, no "and that's why…", no advice.
  4. The share line: who the viewer will tag ("Tag your Mitch").

If 3 is a lecture or 4 is nobody, change the premise.

Beat map (20-28 s, 9-11 lines)

TimeBeatRules
frame 0The hook: POV card + mid-conversation wideNo title card, no "hey guys", no establishing shot. The first word is already the scene.
0-2 sThe 2-second catch: line 1 is a complete, absurd-but-true-to-life claim9 words max, ends by 2.0 s with the final voices (by ~2.4 s with say stand-ins). It must be funny or odd on its own, without the reply ("I'm gonna be a millionaire at ninety-four.").
2-4 sMaya's flat reply (1-4 words)Repeats or names the absurd part ("Ninety-four."). This is the second hook: it tells the viewer the show is on their side.
4-12 sThe double act escalatesMitch's position, Ollie's panic, each line shorter or sillier than the last; one "never?"/"wait" beat.
~10-14 sFirst laughA line that would work as a quote card.
~14-20 sSomeone turns to Maya ("Maya, back me up.")The set-up for her.
~20-24 sThe laugh ("punchline": true)Maya's deadpan 3-5 words, usually. The rimshot and reaction close-up come from the engine.
last lineThe thought: Maya, "punchline": false, pauseBeforeMs: 300, wideThe train stops: the scene cuts to a crew-train-station* set. It reframes the whole talk.
+1.4 sQuiet hold (timing.tailMs: 1400), loop to frame 1No sign-off, no "follow for part 2".

Some episodes may end on the laugh instead (the log tracks the mix; aim for about 1 in 4).

The stop is the signature. The window scrolls the whole episode and stops when Maya lands the thought. Always cut to the station for the last line (on the train sets).

Line rules

  • Under 12 words; most under 8. Contractions, fragments, interruptions ("—").
  • Each says it their way: Mitch "mate", "hear me out", "legend"; Ollie "what if", lists of three, numbers; Maya questions and repeats.
  • Spell numbers the way they're said ("ninety-four"); one picturable number per episode at most.
  • No narrator, no "fun fact", no moral. pnpm check lists phrases that read model-written: fix every ai-wording warning.
  • Cold-viewer read: say each line aloud as if you'd never met them. Every reference must land.

The three places

Pick the place the topic lives in; rotate so no place runs three episodes in a row. Each place has its own seating, so copy that place's cast block exactly (marks differ per set).

SetUse forWho is whereThe ending
crew-train-baycommute, money, plans, phones, "we should…"Mitch stands (far-left, lean-in); Ollie (left) and Maya (right) sit facing across a tablecut to crew-train-station / -station-carts: the train stops
crew-meeting-roomwork: meetings, emails, deadlines, bosses, "quick call"all sit: Ollie (left, facing right) and Maya (right, facing left) side-on at the table's ends; Mitch at the back facing the camera (center)hold wide; coworkers keep walking past the glass
crew-backyard-bbqweekends, family, neighbours, sport, food, plans that never happenall stand: Mitch behind the grill (far-left), Ollie (left) and Maya (right) with drinkshold wide under the string lights

Cast blocks (seated needs the seat at that mark):

// crew-meeting-room
{ "id": "mitch", "character": "crew-mitch", "mark": "center", "seated": true },
{ "id": "ollie", "character": "crew-ollie", "mark": "left", "seated": true, "facing": "right" },
{ "id": "maya",  "character": "crew-maya",  "mark": "right", "seated": true, "facing": "left" }
// crew-backyard-bbq
{ "id": "mitch", "character": "crew-mitch", "mark": "far-left", "facing": "right" },
{ "id": "ollie", "character": "crew-ollie", "mark": "left", "facing": "right", "holding": { "prop": "soda-can" } },
{ "id": "maya",  "character": "crew-maya",  "mark": "right", "facing": "left", "holding": { "prop": "cup" } }

Anyone can take the back seat in the meeting room (Maya facing the camera suits a closing thought); swap the marks in the skit. The "train stops" signature belongs to the train. New places are new crew-* sets (isolation rule), never edits to another kit.

Staging (copy the reference skit's shape)

  • "set": one of the three places, "style": "snappy", "coldOpen": "pov", the cast block for that place (above; the reference skits show the train's). On the train, the last scene: "set": "crew-train-station-carts" (or crew-train-station) with "transition": { "type": "cut" }.
  • Per-line shot: line 1 wide; singles medium on the speaker; close only for the first laugh's speaker or a shock, and only on a line longer than ~1 s (closeup-hold warns); a wide every 3-4 lines and on any line addressed to the group; the last line wide.
  • The closeup-budget warnings ("two plain lines in a row") are expected: the channel cuts to the speaker on every line. Every other warning gets fixed.
  • timing: leadInMs 200, gapMs 120, tailMs 1400.

Running gags (reward return viewers; never required)

  • The coffee carts on the platform: one cart in Ep 1, two carts and a queue by its end. The platform shot shows the current count. The cart business keeps growing in the background; the friends' plans don't. Update the count in the log when a set variant adds a cart.
  • Monday: they always start things "Monday".
  • Ollie's domains: 43 as of Ep 1; the count only goes up.

Use at most one per episode and never as the main joke.

Packaging

  • POV card (overlay.pov): starts "POV: ", 40 characters or fewer in total, names the situation from the viewer's side ("POV: your friend made a budget").
  • Title (meta.title): 25-50 characters, the absurd claim or the situation, one picturable number at most ("He'll be a millionaire at 94").
  • Caption (meta.description): "Every friend group has …" + "Which one are you? Tag your <Mitch|Ollie|Maya>."
  • Hashtags: relatable, stickfigure, the topic, friends.

Multi-part stories

Default: every episode is a complete story. No "Part 1/5" and no planned arcs at launch.

  • A numbered part is only for a premise that has already outperformed (about 2x the channel's median views after 48 h). The sequel is its own episode: new situation, new lines, its own hook and thought, and a small "Part 2" in the title only. It must work for someone who never saw part 1.
  • Never re-use footage, lines or a recap of the earlier part (a one-line callback is fine). Re-using a previous video's shots is what platforms treat as reused/unoriginal content.
  • Never more than 3 parts. Never end on "follow for part 2": the episode ends on the thought.
  • Continuity lives in the running gags and the characters, not in cliffhangers.

Steps

# 1. Pick the next episode number from the log; id: crew-<slug>
#    public/skits/crew-<slug>/skit.json   (copy the reference's shape)
pnpm voice:say crew-<slug>           # macOS stand-in voices until Gemini voices are wired (B2)
pnpm check crew-<slug>               # fix everything except closeup-budget; length 20-28 s
pnpm render crew-<slug> out/crew-<slug>.mp4

Check the 2-second catch from voice.json: leadInMs + line 1 durationMs must be under ~2,400 ms with say (under 2,000 ms with final voices). Then watch the video, not stills: the POV card readable, captions above the seat row, the laugh's reaction close-up, the cut to the platform, the 1.4 s hold.

Then add the episode to the log (episodes/README.md) and write its script page (episodes/NNN-<slug>.md, the same table as 001-we-launch-monday.md).

Hand-back

  • Video path, length, line-1 end time, and the self-check result (any warnings besides closeup-budget).
  • The premise test (behaviour, two sides, thought, share line) and who is at fault.
  • Which lines are placeholders for the team.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.