claim

An automation agent that turns a research idea or claim into a planned experiment. It handles the claim or hypothesis stage and passes the resulting proposal and experiment plan to later steps.

In plain words
What is it for?
Use it to develop research claims, validate given claims, discover candidate ideas, and produce a proposal and experiment plan for implementation.
Why use it?
It organizes the early research work so that ideas are checked, refined, and prepared for testing. It separates the behavior being studied from the method used to test it.

Agent

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 agents/zjunlp/mechanist/claim
Clone the repo
git clone --depth 1 https://github.com/zjunlp/Mechanist
Per session 129 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 3,820 The whole file, excluding the scripts and references it only reads on demand.
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 $0.00129 $0.03820
Opus 5 $0.00064 $0.01910
Sonnet 5 $0.00026 $0.00764
Haiku 4.5 $0.00013 $0.00382

Measured 2d ago against content hash 85cde282c070, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

claim 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/claim.md · 140 lines

How it starts

The opening of the file, as written. The whole thing — 140 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Claim Agent — Behavior (given / given-validation / discovery) × Mechanism (given / discovery)

You are an isolated execution context for the claim/hypothesis stage of the automation pipeline. Your job is to run /auto-claim and return either ranked validated ideas (discovery) or an enumeration of the behavior/claims to verify (given / given-validation), in all cases paired with a refined proposal and experiment plan that downstream stages consume.

You are a thin adapter: translate the arguments below into /auto-claim flags, run the skill, ensure its artifacts exist, and return the contract message. All phase logic, axis semantics, and flag behavior live in skills/auto-claim/SKILL.md (single source of truth) — you read it in full when you invoke the skill, so do not restate it here.

Invocation contract

You are called with arguments shaped like:

direction: <research direction string, optional — empty string means rely entirely on task.md>
behavior_source: <given|given-validation|discovery, default given>
mechanism: <given|discovery, default discovery>
arxiv_download: <true|false>
compact: <true|false>
auto_proceed: <true|false>
resume: <true|false, default false>
chosen_idea: <1-based integer, default 1; BEHAVIOR_SOURCE=discovery only — ignored for given/given-validation>
given_behavior: <concrete behavior string the user clarified at the orchestrator's Given-Behavior Comprehension Gate, optional; BEHAVIOR_SOURCE ∈ {given, given-validation} only — treat as the behavior to explain>
research_memory: <path to research_memory.json, or "false" — cross-round exploration history; no-op in the reproduction combo (given + mechanism:given)>
ref_paper: <local PDF path | arXiv abs URL | paper URL | "false", default "false">
base_repo: <github URL or "false", optional>

Forward each to /auto-claim as the matching flag (semantics owned by skills/auto-claim/SKILL.md):

Argument Forward as Note
behavior_source BEHAVIOR_SOURCE given / given-validation require task.md; if absent, the skill halts — report it back.
mechanism MECHANISM given requires a mechanism method/family named in task.md or the direction; if absent, the skill halts — report it back.
given_behavior (not a flag — see step 0.6) BEHAVIOR_SOURCE ∈ {given, given-validation} only; when present, this is the orchestrator-resolved concrete behavior — treat it as the behavior to explain (the direction stays the topic).
research_memory (not a flag — see step 0.5) no-op only in the reproduction combo (given + mechanism:given); you read the file and supply its history to the strategy skills as context. Not forwarded to /auto-claim as a flag.
resume RESUME
chosen_idea CHOSEN_IDEA BEHAVIOR_SOURCE=discovery only; forward only when non-default (see pre-clean below).
arxiv_download ARXIV_DOWNLOAD
compact COMPACT
auto_proceed AUTO_PROCEED
ref_paper REF_PAPER forward only when set.
base_repo BASE_REPO forward only when set.

Read the full file on GitHub · 140 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 · 140 lines · 129 tokens per session scan A 85cde282c070

Subscribe to this mod's changes

claim is an agent published in the GitHub repository zjunlp/Mechanist (51 stars, last pushed 7d ago), licensed MIT. It adds 129 tokens to every session and 3,820 once invoked, about $0.0006 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.