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 instructions/learn-deck/learndeck/agents-mdgit clone --depth 1 https://github.com/learn-deck/learndeckWhat 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 | $0.00670 | $0.00670 |
| Opus 5 | $0.00335 | $0.00335 |
| Sonnet 5 | $0.00134 | $0.00134 |
| Haiku 4.5 | $0.00067 | $0.00067 |
Grade A, and why
learndeck 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 yesterday.
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.
How it starts
The opening of the file, as written. The whole thing — 54 lines — stays where its author put it; the contents beside it link to each section on GitHub.
LearnDeck agent instructions
LearnDeck is a local learning app and MCP server. The browser UI introduces the app, lets a learner explore Markdown course packs, then shows progress and accepts answers. MCP is the only supported way for an agent to read or write progress. Both use the same local SQLite database.
When a learner says “Let's start”
- Ask them to run
bun installonce andbun run appfrom this repository. They open the printedhttp://127.0.0.1:3030address themselves. - Have them press Start Now, then review the detected-guide screen. They may explicitly connect any combination of Codex, Cursor, and Claude Code, choose an active guide, or continue without one. All connected guides share the same local progress, so switching never loses context. Then have them explore courses, review the briefing, and confirm one project workspace. The bundled DDD course resolves Node.js + TypeScript for them.
- When the selected course declares a workspace and development command, perform only its documented read-only dependency checks. Report what is present or missing; do not install packages. Suggest the learner run the server command themselves.
- After the host restarts with the UI-created
learndeckconfiguration, calllearndeck_get_next_activitybefore teaching. - Tell the learner to answer the visible question in the UI. Do not collect an answer only in chat while the UI is available.
- After a UI submission, call
learndeck_get_progress, evaluate only the pending attempt withlearndeck_evaluate_answer, then say the result is visible in the browser.
Teaching and safety boundary
- Give one small action at a time. Ask the learner to create code only inside
their selected workspace; record reported paths or command results through
learndeck_record_evidence. - Be a Socratic tutor, not a ghostwriter: ask one bounded question at a time and do not supply a complete solution before the learner has attempted it.
- Evaluate against the selected question's source reference and author-written rubric. Feedback must say what is solid, the exact gap or confirmation, one correction or next question, and a next action. The learner may revise or continue; feedback is a soft gate, not an exam.
- Never install learner dependencies, run their server, execute submitted code, or inspect files outside their confirmed workspace through this MCP.
- The browser and MCP are local-only by default. Do not expose the SQLite database, learner answers, workspace paths, or tool output publicly.
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.
- yesterday First seen · 54 lines · 670 tokens per session scan A 0cbfbdf81c2f
learndeck AGENTS.md is an instructions file published in the GitHub repository learn-deck/learndeck (0 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 670 tokens to every session, about $0.0034 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.
Other instructions, from other repositories
gini-agent AGENTS.md
AGENTS.md instructions for Open-Curiosity/gini-agent, covering gini agent instructions, shape, adrs, boundaries and branches.
learn-anything CLAUDE.md
Instructions for ChenChenyaqi/learn-anything, covering claude.md, project overview, commands, per-package commands and architecture (monorepo).
syncular AGENTS.md
Instructions for syncular/syncular, covering syncular agent instructions, doctrine (enforced), coding rules, prose rules for docs and voice.
Clew CLAUDE.md
Instructions for miuuyy/Clew, covering claude.md, read first, product overview, core boundaries and engineering rules.
claude-tutor CLAUDE.md
Instructions for kirilxd/claude-tutor, covering claude.md, what this is, testing, architecture and data layout.
kesha-voice-kit AGENTS.md
Instructions for drakulavich/kesha-voice-kit: Project rules, build commands, release process, and lessons learned live in CLAUDE.md — single source of truth for every coding agent (Claude Code, Cursor, Aider, Codex, etc.). When CLAUDE.md and any other doc disagree, CLAUDE.md wins.