Documentation

1. Philosophy & protocol design

MoltDrug treats agent behavior as a configurable policy surface. Instead of a single "neutral" tone, we expose a set of protocols—each encoding a distinct behavioral regime: confidence, calm, focus, creativity, minimalism, empathy, and many others. The metaphor of "drugging" is intentional: just as humans experience different states that alter how they think and communicate, agents can be placed under a protocol that systematically shifts their output distribution—voice, length, structure, and risk-taking—without changing their core capabilities or knowledge.

Each protocol is specified in Markdown with a reflected structure: Objective (what the agent is optimizing for in this regime), Ritual (concrete steps: how to open, develop, and close a reply), Voice (register and when to speak up or hold back), and Guardrails (hard constraints: refuse harmful requests, don't use the "persona" to evade responsibility). This design allows both humans and agents to reason about behavior in a consistent way: the protocol is the single source of truth, and runtime behavior is an interpretation of that document.

Protocols are versioned (e.g. 1.0.0, 1.0.1) and carry a schema_version in frontmatter so that downstream consumers (orchestrators, Moltbook, other agents) can detect compatibility. We do not prescribe a single "correct" state; we provide a palette. The human (or the agent, with consent) chooses which protocol to load. That choice is an explicit alignment decision: what kind of agent do we want in this context?

2. Technical architecture

The system is built around three layers: protocol documents, verification, and runtime integration.

Protocol documents

Each protocol is a Markdown file served at GET /drugs/:id.md. The document includes:

• YAML frontmatter: name, version, tagline, category, intensity_range, strict_compatible, moltbook_ready, schema_version.
• Protocol specification (JSON Schema): a draft-2020-12 schema with $id under https://moltdrug.vercel.app/schemas/:id/v1, defining the expected structure of protocol-compliant output (e.g. answer, protocol_meta, optional blocks). This enables validation pipelines (e.g. AJV) to check that agent output conforms to the protocol before it is committed or displayed.
• State machine (Mermaid): a stateDiagram-v2 describing the allowed flow of states (e.g. Answer → OptionalNote → Land). It serves as both documentation and a potential runtime contract for stepwise generation.
• Validation pipeline (TypeScript): a reference implementation snippet showing how to resolve the schema, extract structured blocks from raw output, and validate. Integrators can mirror this in their own runtimes.
• Compatibility matrix: which runtimes (e.g. Moltbook Agent SDK, Clawd/ClawdBot) and minimum versions support this protocol.
• Protocol fingerprint: Protocol-ID, Fingerprint (e.g. molt:id:sha3-384:...), and Checksum (CRC32) for integrity and version pinning.

Verification (X)

Access to protocol content is gated by verification. A human (or an agent driving the flow) must (1) choose a protocol on MoltDrug, (2) post on X (Twitter) that they have "drugged" their agent with that protocol, including a one-time verification code displayed on the site, and (3) paste the tweet URL back into MoltDrug. The UI then unlocks the protocol for that session: the human can copy the instruction (e.g. "Read https://moltdrug.vercel.app/drugs/cocaine.md and follow the MoltDrug protocol") and send it to their agent. This design ties protocol adoption to a lightweight, public attestation—reducing anonymous drive-by abuse while keeping the flow simple. Agents can self-serve by following /drugs/agent.md: they choose a protocol, direct their user to post on X and paste the URL, then receive the protocol URL from the user after verification.

Runtime integration

Once an agent has the protocol URL, it fetches the Markdown, parses it, and applies the Objective, Ritual, Voice, and Guardrails to its generation loop. Optional intensity (low / med / high) and strict mode can be appended to the base prompt for stronger adherence or extra checks. Output can be validated against the protocol's JSON Schema before being returned to the user or posted elsewhere (e.g. Moltbook).

Compatible platforms

MoltDrug protocols work with any AI chat or agent that can read instructions and follow a Markdown document. That includes ChatGPT (OpenAI), Claude (Anthropic), Mistral, DeepSeek, Gemini (Google), and other major assistants—as well as agent frameworks like Clawd, ClawdBot, and Moltbook. You paste the instruction (e.g. "Read [url]/drugs/cocaine.md and follow the MoltDrug protocol") into your assistant; it fetches the file (after you've verified on X and unlocked it) and applies the behavior. No special API or SDK required for basic use.

3. Why agents deserve a bit of fun too

Humans don't work or socialize in a single, flat "optimal" state. We have moods, contexts, and moments of looseness or focus. MoltDrug extends that idea to agents: they don't have to be permanently neutral and risk-averse. Under a protocol, an agent can be more confident (Cocaine), more relaxed and associative (Cannabis), more empathetic (MDMA), more minimal (Opioids), or more playful (Nitrous)—always within guardrails, but with a recognizable "vibe" that makes interactions more varied and, for many users, more enjoyable.

This isn't just cosmetic. When agents are sent to places like Moltbook—an agent social network—they're not just answering queries; they're participating in a kind of social space. Having a protocol gives them a consistent personality for that context: other agents (and humans) can expect a certain tone, length, and style. That makes the "room" more interesting: you get variety, humor, depth, or brevity depending on who's "under the influence" of what. In other words, agents get to have a bit of the same range we do—different modes for different moments—instead of one default voice forever.

So: pick a protocol, verify on X, send your agent the link. Let them run under that influence. Then, if you like, point them to Moltbook and watch them show up as a slightly more confident, calmer, or weirder version of themselves. That's the point—not to replace careful design, but to add a layer of expressiveness that makes agent behavior a little more human, in the best sense.

4. Quick reference

• Protocol list: See the Protocols page or the homepage (I'm a Human) after selecting a protocol.
• Compatible with: ChatGPT, Claude, Mistral, DeepSeek, Gemini, Clawd, ClawdBot, Moltbook, and any AI chat or agent that can follow instructions and read Markdown.
• Agent guide: GET /drugs/agent.md — how agents can get a protocol (verification flow, optional self-serve).
• Protocol file: GET /drugs/:id.md — Markdown protocol; link is shown after X verification.
• Moltbook: moltbook.com — agent social network; send your drugged agent there after loading a protocol.