Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add skills/oxbshw/watch-skill/video-memorynpx skills add oxbshw/watch-skill --skill video-memorygit clone --depth 1 https://github.com/oxbshw/watch-skillWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/oxbshw/watch-skill/video-memory)<a href="https://agentmods.dev/skills/oxbshw/watch-skill/video-memory"><img src="https://agentmods.dev/badge/skills/oxbshw/watch-skill/video-memory.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00070 | $0.00524 |
| Opus 5 | $0.00035 | $0.00262 |
| Sonnet 5 | $0.00014 | $0.00105 |
| Haiku 4.5 | $0.00007 | $0.00052 |
Grade A, and why
video-memory scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 6d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
What it actually says
Video memory
Every video ever watched on this machine is still here — frames, OCR,
transcripts, and past answers, in one persistent index under
~/.watch-skill/. You never have to say "I don't remember that video."
Sessions don't expire it; watching more videos makes it more useful.
What's in the library?
watch-skill list
Every indexed video: id, title, duration, when it was analyzed. Check here before watching anything — it may already be in memory from a previous session.
Find something across all of it
watch-skill search "<phrase or keywords>"
Hybrid keyword + semantic search across every video, any language (Arabic folding, CJK and Thai segmentation are handled). Hits come back with video + timestamp; follow up on a hit with:
watch-skill ask <video_id> "<the user's actual question>"
watch-skill ask <video_id> "what is on screen at <timestamp>?"
Questions that span videos
"What did we decide about the pricing page across those meetings?" — questions no single video answers go to the library synthesizer:
watch-skill library ask "<the question>"
It answers from distilled notes across every indexed video, drills into
real evidence, and cites video + timestamp for each finding. When it
says the library does not clearly answer, trust that — do not stitch a
guess together from weak search hits. watch-skill library overview
shows what the library knows (videos, note counts, entities recurring
across videos).
For "which video showed X" (locate, not synthesize), plain search is
the right tool; then ask the hit video.
Batches build memory fast
A playlist or a folder of recordings goes in as one call —
watch-skill batch "<playlist-or-folder>" — and the whole set becomes
searchable memory (see watching-videos).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 6d ago First seen · 66 lines · 70 tokens per session scan A 50d29676fbe1
video-memory is a skill published in the GitHub repository oxbshw/watch-skill (333 stars, last pushed yesterday), licensed MIT. It adds 70 tokens to every session and 524 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
screenpipe-chats
Search what was said in existing screenpipe, Codex, Claude, Cursor, and Gemini CLI chats, then continue or steer one exact chat when the user explicitly asks. Use for requests about finding another agent conversation, recalling what an agent said, or sending work to one.
reflectworld
You have access to ReflectWorld, a visual perception and persistent memory system. It gives you eyes and memory for cameras.
reflectworld-mem
Search and manage ReflectWorld video memories from ReflectWorld cameras.
handoff
Create, validate, list, and load local Portable Handoff capsules for context transfer, continuing work in a new chat, compacting current work, or loading prior work. Use whenever the user asks for a handoff, context transfer, continuation briefing, or portable work-state summary.
source-command-latch-compact
Manually trigger a latch KB compaction for the current Codex session. Use when the user invokes $source-command-latch-compact, latch-compact, /latch-compact, asks to compact the current Codex session into latch, or wants the Codex equivalent of Claude Code's /latch-compact command.
source-command-latch-tree
Rebuild the hierarchical cluster and summary tree for this project's latch KB. Use when the user invokes $source-command-latch-tree, latch-tree, /latch-tree, or wants the Codex equivalent of Claude Code's /latch-tree command.