feat(MAX-568): add /memory OSS demo-first reel pack

content/reels/MAX-568/README.md

@@ -0,0 +1,26 @@
+# MAX-568, /memory OSS demo-first reel pack
+
+This package ships the requested 5 short-form reel variants for `/memory` in the correct repo, `maxtechera/memory`.
+
+## Included assets
+- `scripts.md`, the full hook and script library
+- `videos/*.mp4`, 5 rendered 9:16 reel proof assets
+- `thumbnails/*.png`, 5 thumbnail images for Linear and social upload
+- `proof-board.png`, one combined visual proof image
+- `render_variants.sh`, reproducible local renderer for the pack
+
+## Variant map
+1. Problem agitate solve
+2. Contrarian
+3. Specific number
+4. Insider reveal
+5. Testimonial
+
+## CTA map
+- Primary: IG DM keyword `MEMORY` for the free install guide
+- Secondary: comment `install` for the repo link reply
+- Tertiary: link in bio → `github.com/maxtechera/memory`
+
+## Note on ManyChat
+The ticket contract references `keyword "MEMORY" → flow_id TBD_MAX_PROVIDES`.
+No concrete flow id was available in the workspace, so the assets keep the DM keyword language but do not hardcode a missing flow id.

content/reels/MAX-568/render-text/01-problem-agitate-solve-cta.txt

@@ -0,0 +1 @@
+DM MEMORY for the free install guide

content/reels/MAX-568/render-text/01-problem-agitate-solve-headline.txt

@@ -0,0 +1 @@
+Your AI forgets everything.

content/reels/MAX-568/render-text/01-problem-agitate-solve-scene1.txt

@@ -0,0 +1,3 @@
+Open a fresh session.
+The repo is gone.
+The last decision is gone.

content/reels/MAX-568/render-text/01-problem-agitate-solve-scene2.txt

@@ -0,0 +1,3 @@
+So you re-explain the brief,
+restate the task,
+and burn tokens catching up.

content/reels/MAX-568/render-text/01-problem-agitate-solve-scene3.txt

@@ -0,0 +1,3 @@
+Install /memory once.
+Next session resumes with repo,
+last decision, and current focus.

content/reels/MAX-568/render-text/02-contrarian-cta.txt

@@ -0,0 +1 @@
+Comment install for the repo link

content/reels/MAX-568/render-text/02-contrarian-headline.txt

@@ -0,0 +1 @@
+The problem is not your prompt.

content/reels/MAX-568/render-text/02-contrarian-scene1.txt

@@ -0,0 +1,3 @@
+Longer prompts do not fix
+a stateless agent.
+They just make the reset slower.

content/reels/MAX-568/render-text/02-contrarian-scene2.txt

@@ -0,0 +1,3 @@
+The real gap is continuity.
+Your agent needs memory,
+not a bigger context dump.

content/reels/MAX-568/render-text/02-contrarian-scene3.txt

@@ -0,0 +1,3 @@
+HOT stays loaded.
+WARM loads by topic.
+COLD stays searchable.

content/reels/MAX-568/render-text/03-specific-number-cta.txt

@@ -0,0 +1 @@
+Link in bio or DM MEMORY

content/reels/MAX-568/render-text/03-specific-number-headline.txt

@@ -0,0 +1 @@
+Three tiers beat one giant prompt.

content/reels/MAX-568/render-text/03-specific-number-scene1.txt

@@ -0,0 +1,2 @@
+HOT stores the active state
+you need every session.

content/reels/MAX-568/render-text/03-specific-number-scene2.txt

@@ -0,0 +1,2 @@
+WARM keeps reusable context
+ready by topic or project.

content/reels/MAX-568/render-text/03-specific-number-scene3.txt

@@ -0,0 +1,2 @@
+COLD keeps deep history searchable,
+so continuity survives compaction.

content/reels/MAX-568/render-text/04-insider-reveal-cta.txt

@@ -0,0 +1 @@
+DM MEMORY for the install guide

content/reels/MAX-568/render-text/04-insider-reveal-headline.txt

@@ -0,0 +1 @@
+The pre-compact flush is the secret.

content/reels/MAX-568/render-text/04-insider-reveal-scene1.txt

@@ -0,0 +1,2 @@
+Most memory setups fail
+right before the context resets.

content/reels/MAX-568/render-text/04-insider-reveal-scene2.txt

@@ -0,0 +1,3 @@
+/memory writes session state
+before the reply, then flushes again
+before compaction lands.

content/reels/MAX-568/render-text/04-insider-reveal-scene3.txt

@@ -0,0 +1,3 @@
+That is why the next session
+can actually pick up where
+the last one stopped.

content/reels/MAX-568/render-text/05-testimonial-cta.txt

@@ -0,0 +1 @@
+Comment install or grab the repo

content/reels/MAX-568/render-text/05-testimonial-headline.txt

@@ -0,0 +1 @@
+The moment my agent stopped forgetting, work sped up.

content/reels/MAX-568/render-text/05-testimonial-scene1.txt

@@ -0,0 +1,2 @@
+Every morning used to start
+with the same long re-brief.

content/reels/MAX-568/render-text/05-testimonial-scene2.txt

@@ -0,0 +1,3 @@
+After /memory, new sessions
+already knew the repo,
+last decision, and next step.

content/reels/MAX-568/render-text/05-testimonial-scene3.txt

@@ -0,0 +1,3 @@
+The work felt continuous,
+not disposable.
+That changed everything.

content/reels/MAX-568/render_variants.sh

@@ -0,0 +1,101 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+ROOT="$(cd "$(dirname "$0")" && pwd)"
+OUT="$ROOT/videos"
+THUMBS="$ROOT/thumbnails"
+TXT="$ROOT/render-text"
+FONT=$(fc-match -f '%{file}\n' 'DejaVu Sans' | head -1)
+mkdir -p "$OUT" "$THUMBS" "$TXT"
+
+render_variant() {
+  local slug="$1"
+  local kicker="$2"
+  local headline="$3"
+  local scene1="$4"
+  local scene2="$5"
+  local scene3="$6"
+  local cta="$7"
+
+  printf '%s\n' "$headline" > "$TXT/$slug-headline.txt"
+  printf '%s\n' "$scene1" > "$TXT/$slug-scene1.txt"
+  printf '%s\n' "$scene2" > "$TXT/$slug-scene2.txt"
+  printf '%s\n' "$scene3" > "$TXT/$slug-scene3.txt"
+  printf '%s\n' "$cta" > "$TXT/$slug-cta.txt"
+
+  ffmpeg -y -f lavfi -i color=c=0x08111f:s=1080x1920:d=12 -vf "
+    drawbox=x=60:y=60:w=960:h=180:color=0x10233f:t=fill,
+    drawbox=x=60:y=300:w=960:h=1220:color=0x0d1b31:t=fill,
+    drawbox=x=60:y=1580:w=960:h=210:color=0x10233f:t=fill,
+    drawtext=fontfile=$FONT:text='${kicker}':fontcolor=0x7dd3fc:fontsize=34:x=90:y=110,
+    drawtext=fontfile=$FONT:textfile=$TXT/$slug-headline.txt:fontcolor=white:fontsize=68:x=90:y=180:line_spacing=12,
+    drawtext=fontfile=$FONT:text='01':enable='lt(t,4)':fontcolor=0x7dd3fc:fontsize=40:x=90:y=380,
+    drawtext=fontfile=$FONT:textfile=$TXT/$slug-scene1.txt:enable='lt(t,4)':fontcolor=0xe2e8f0:fontsize=44:x=160:y=382:line_spacing=16,
+    drawtext=fontfile=$FONT:text='02':enable='between(t,4,8)':fontcolor=0x7dd3fc:fontsize=40:x=90:y=380,
+    drawtext=fontfile=$FONT:textfile=$TXT/$slug-scene2.txt:enable='between(t,4,8)':fontcolor=0xe2e8f0:fontsize=44:x=160:y=382:line_spacing=16,
+    drawtext=fontfile=$FONT:text='03':enable='gte(t,8)':fontcolor=0x7dd3fc:fontsize=40:x=90:y=380,
+    drawtext=fontfile=$FONT:textfile=$TXT/$slug-scene3.txt:enable='gte(t,8)':fontcolor=0xe2e8f0:fontsize=44:x=160:y=382:line_spacing=16,
+    drawtext=fontfile=$FONT:textfile=$TXT/$slug-cta.txt:fontcolor=0x7dd3fc:fontsize=40:x=90:y=1650:line_spacing=14,
+    drawtext=fontfile=$FONT:text='github.com/maxtechera/memory':fontcolor=white:fontsize=30:x=90:y=1765,
+    drawtext=fontfile=$FONT:text='OSS demo-first reel pack':fontcolor=0x94a3b8:fontsize=28:x=90:y=1820
+  " -c:v libx264 -pix_fmt yuv420p -r 30 "$OUT/$slug.mp4"
+
+  ffmpeg -y -i "$OUT/$slug.mp4" -vf "select=eq(n\,0)" -vframes 1 -update 1 "$THUMBS/$slug.png"
+}
+
+render_variant \
+  "01-problem-agitate-solve" \
+  "MAX-568 /memory" \
+  "Your AI forgets everything." \
+  $'Open a fresh session.\nThe repo is gone.\nThe last decision is gone.' \
+  $'So you re-explain the brief,\nrestate the task,\nand burn tokens catching up.' \
+  $'Install /memory once.\nNext session resumes with repo,\nlast decision, and current focus.' \
+  "DM MEMORY for the free install guide"
+
+render_variant \
+  "02-contrarian" \
+  "MAX-568 /memory" \
+  "The problem is not your prompt." \
+  $'Longer prompts do not fix\na stateless agent.\nThey just make the reset slower.' \
+  $'The real gap is continuity.\nYour agent needs memory,\nnot a bigger context dump.' \
+  $'HOT stays loaded.\nWARM loads by topic.\nCOLD stays searchable.' \
+  "Comment install for the repo link"
+
+render_variant \
+  "03-specific-number" \
+  "MAX-568 /memory" \
+  "Three tiers beat one giant prompt." \
+  $'HOT stores the active state\nyou need every session.' \
+  $'WARM keeps reusable context\nready by topic or project.' \
+  $'COLD keeps deep history searchable,\nso continuity survives compaction.' \
+  "Link in bio or DM MEMORY"
+
+render_variant \
+  "04-insider-reveal" \
+  "MAX-568 /memory" \
+  "The pre-compact flush is the secret." \
+  $'Most memory setups fail\nright before the context resets.' \
+  $'/memory writes session state\nbefore the reply, then flushes again\nbefore compaction lands.' \
+  $'That is why the next session\ncan actually pick up where\nthe last one stopped.' \
+  "DM MEMORY for the install guide"
+
+render_variant \
+  "05-testimonial" \
+  "MAX-568 /memory" \
+  "The moment my agent stopped forgetting, work sped up." \
+  $'Every morning used to start\nwith the same long re-brief.' \
+  $'After /memory, new sessions\nalready knew the repo,\nlast decision, and next step.' \
+  $'The work felt continuous,\nnot disposable.\nThat changed everything.' \
+  "Comment install or grab the repo"
+
+ffmpeg -y \
+  -i "$THUMBS/01-problem-agitate-solve.png" \
+  -i "$THUMBS/02-contrarian.png" \
+  -i "$THUMBS/03-specific-number.png" \
+  -i "$THUMBS/04-insider-reveal.png" \
+  -i "$THUMBS/05-testimonial.png" \
+  -filter_complex "[0:v][1:v][2:v][3:v][4:v]hstack=inputs=5,scale=1620:576" \
+  -frames:v 1 -update 1 "$ROOT/proof-board.png"
+
+echo "Rendered videos to $OUT"
+echo "Rendered thumbnails to $THUMBS"

content/reels/MAX-568/scripts.md

@@ -0,0 +1,75 @@
+# MAX-568, /memory OSS demo-first reel pack
+
+## Goal
+Create a 5-variant short-form reel package for `/memory` that proves the open-source memory flow visually and funnels viewers toward the free install guide and repo.
+
+## Audience
+AI and LLM developers, Claude Code users, and indie hackers tired of re-explaining project context every session.
+
+## Core angle
+Your AI forgets everything. `/memory` fixes that with a 3-tier HOT, WARM, and COLD memory system.
+
+## CTA map
+- Primary: DM `MEMORY` for the free install guide
+- Secondary: comment `install` for the repo link reply
+- Tertiary: link in bio → `github.com/maxtechera/memory`
+
+## Variant 1, Problem agitate solve
+### Hooks
+1. Your AI forgets everything.
+2. Same repo, same task, still repeating the brief.
+3. If every session starts from zero, your agent is not helping.
+
+### Script
+Show the failure first: fresh session, lost context, repeated briefing.
+Then show the fix: install `/memory`, run setup once, let hooks persist state.
+Land the promise: next session resumes with the repo, last decision, and current focus already loaded.
+CTA: DM `MEMORY` for the free install guide.
+
+## Variant 2, Contrarian
+### Hooks
+1. The issue is not your prompting. It is missing memory.
+2. Longer prompts will not save a stateless agent.
+3. Stop writing context dumps. Give your agent memory.
+
+### Script
+Open with the common bad advice, then flip it.
+Better prompts are not enough when every session forgets.
+`/memory` fixes the continuity layer with HOT state in-context, WARM notes by topic, and COLD search for deep recall.
+CTA: Comment `install` and get the repo link.
+
+## Variant 3, Specific number
+### Hooks
+1. Three memory tiers beat one giant prompt.
+2. One setup, three layers, zero re-briefing.
+3. HOT, WARM, COLD is the stack your agent actually needed.
+
+### Script
+Lead with the number-driven hook.
+Then walk through the three-tier model in plain language.
+End on the practical result: less token waste, cleaner resumes, better continuity after compaction.
+CTA: Link in bio or DM `MEMORY`.
+
+## Variant 4, Insider reveal
+### Hooks
+1. The real trick is the write-ahead memory step.
+2. The pre-compact flush is why this survives resets.
+3. Most memory setups fail before the next session starts.
+
+### Script
+Reveal the hidden mechanism.
+`/memory` writes session state before the reply and flushes again before compaction, so decisions survive resets.
+Tie that back to the demo: that is why the second session can pick up immediately.
+CTA: DM `MEMORY` for the install guide.
+
+## Variant 5, Testimonial
+### Hooks
+1. The moment my agent stopped forgetting, everything sped up.
+2. `/memory` made my sessions feel continuous.
+3. This is the first memory setup that fit my actual workflow.
+
+### Script
+Tell the story from the user point of view.
+Before: every morning started with a long re-brief.
+After `/memory`: new sessions already knew the repo, last decision, and current task, so the work felt continuous instead of disposable.
+CTA: Comment `install` or grab the repo from the link in bio.