vidtheque AGENTS.md

Repository instructions for vidtheque, a tool that turns followed videos into timestamped, receipt-backed knowledge for people and their AI agents.

In plain words
What is it for?
Use them when planning work, changing the MCP or worker components, consulting project evidence, or checking boundaries such as the required HTTP-only connection.
Why use it?
They define which design, product, research, and roadmap documents take priority, while recording rules learned from past incidents. This helps agents avoid conflicting or outdated guidance.

Instructions file for CodexOpenCode

Install

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.

agentmods
npx agentmods add instructions/t0msilver/vidtheque/agents-md
Clone the repo
git clone --depth 1 https://github.com/T0mSIlver/vidtheque

Made for: Codex, OpenCode.

Per session 990 This file is loaded in full into every session.
When invoked 990 The same file — it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5 $0.00990 $0.00990
Opus 5 $0.00495 $0.00495
Sonnet 5 $0.00198 $0.00198
Haiku 4.5 $0.00099 $0.00099

Measured 2d ago against content hash 307ec661622c, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

vidtheque AGENTS.md 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 2d 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.

AGENTS.md · 76 lines

How it starts

The opening of the file, as written. The whole thing — 76 lines — stays where its author put it; the contents beside it link to each section on GitHub.

AGENTS.md — how to work in this repo

vidtheque turns the videos you follow into timestamped, receipt-backed knowledge for you and your agents. Owner: Tom Vaucourt. MIT. This file is the working conventions, kept short because it loads into every session; the product itself is explained in README.md and docs/README.md.

Ground truth, in order

  1. docs/design/*.mdthe contracts (DECISIONS.md outranks the rest). If implementation must diverge, amend the contract in the same commit and say why. positioning.md is LOCKED: all public-facing copy follows it.
  2. research/ — append-only evidence (index: research/README.md; founding decisions: research/HANDOFF-2026-08-08.md). Add dated sections; never rewrite or delete.
  3. DESIGN.md + PRODUCT.md — the visual contract for the web surfaces.
  4. This file — conventions only. Any doc above wins.

docs/ROADMAP.md is the only list of open work — a backlog inside a dated document is a snapshot and goes stale silently. docs/LESSONS.md holds the operational rules an incident or a launch paid for.

Invariants — never change these without Tom

  • mcp/worker/ is HTTP only. No Python imports across the boundary, not even in tests. The worker is stateless inference; all state lives in mcp/.
  • Two time axes, never overloaded: published_* picks videos, offset_* picks positions inside one.
  • all means all — a filter that can't apply to a leg prints a note:, never silently narrows.
  • Token discipline: middle-truncation with a documented 0 opt-out, pagination hints printed in payloads, double caps (items AND chars), expensive paths bounded independently of limit, server-side clamps — never prompt-only limits.
  • Frames by authenticated URL by default; inline base64 is the opt-in, with the correct mimeType.
  • has_more over exact totals; relevance-first ordering, order explicit.

Working here

  • Python 3.12, uv workspace (mcp/, worker/). make test is CPU-only and must stay green — tests never download models or need a GPU. make bench is manual, Tom's box only.
  • Deterministic builds: committed uv.lock, digest-pinned CUDA base image. Dependency bumps are deliberate commits, not side effects.
  • deploy/.env.example is the document of record for every env var — a new env without an entry there is a bug.
  • No self-hosted CI runners, ever (public repo, drive-by PR risk).
  • Tom's box (3090, llama.cpp VRAM lease via GPU_ACQUIRE_CMD hooks) is context for deploy work, not a shipped default — see the founding handoff.

Read the full file on GitHub · 76 lines

Changes

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.

  1. 2d ago First seen · 76 lines · 990 tokens per session scan A 307ec661622c

Subscribe to this mod's changes

vidtheque AGENTS.md is an instructions file published in the GitHub repository T0mSIlver/vidtheque (2 stars, last pushed 2d ago), licensed MIT. It adds 990 tokens to every session, about $0.0049 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 2026-08-31.