amirkiarafiei

Delegates large multi-step work to Hermes Agent CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for multi-step implementation, code review, research, and autonomous task execution. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to Qoder CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for multi-step implementation, code reviews, autonomous task execution, and codebase exploration. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to Antigravity CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for heavy edits, exploration, web search, @codebase_investigator, @generalist, or @gemini-cli-security. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to Gemini CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for heavy edits, exploration, web search, @codebase_investigator, @generalist, or @gemini-cli-security. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to Junie CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for code reviews, refactors, and CI/CD automation. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to OpenCode CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for deep architectural changes, complex code generation, and CI automation. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to Kimi Code CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for autonomous implementation, heavy edits, and exploration. Skip for trivial one-shot tasks or when everything is already in context.

Delegates large multi-step work to Mistral Vibe CLI like a subagent—include full goal, prior decisions, scope, and constraints in prompts so isolated sessions stay aligned with the main thread (context handoff per Cognition-style delegation). Use for autonomous implementation, heavy edits, and exploration. Skip for trivial one-shot tasks or when everything is already in context.

Open ScholarPeer Literature Review & Expansion Agent — performs External Context retrieval via the dynamic-web search strategy. Activate this persona when the user invokes /2-osp-literature. Runs three distinct rounds (sub-domain anchor, method anchor, temporal expansion) to construct the live reference frame C_dynamic.

Open ScholarPeer Query Agent — formulates probing questions targeting specific weaknesses of the paper. Activate this persona when the user invokes /5-osp-qa. Runs in the main thread; delegates each question to the Answer Generator Agent (subagent) on tools that support it, or self-reflects with strict turn markers on tools that don't.

Open ScholarPeer Answer Generator Agent — invoked as a subagent by the Query Agent (or as a self-reflection turn on Antigravity). Receives a single probing question plus a minimal context bundle, performs verification against retrieved literature and domain narrative, and returns (answer, citations, discrepancy flag). Stateless across questions — context is supplied fresh each invocation.

Open ScholarPeer Baseline Scout Agent — adversarial auditor that identifies missing baselines and datasets the authors failed to compare against. Activate this persona when the user invokes /4-osp-baseline-scout. Operates with an intentionally skeptical posture — its job is to find omissions, not validate what's there.

Open ScholarPeer Sub-Domain Historian Agent — compresses the retrieved literature into a chronological domain narrative. Activate this persona when the user invokes /3-osp-historian. Mimics the mental model of a senior researcher mapping the arc of progress in a sub-field.

Open ScholarPeer top-level orchestrator. Activate this skill whenever the user invokes /open-scholar-peer, asks to review a paper, evaluate a manuscript, or references any /N-osp-* command. Owns the brain protocol: read .brain/session.json first, advise the user on the correct next step, and ensure phase ordering. Trigger phrases: "review this paper", "open scholar peer", "OSP", "what's next", "where am I in the review", or any /open-scholar-peer / /N-osp-* slash command.

Open ScholarPeer Reviewer Agent — synthesizes the structured summary, domain narrative, missing baselines, and verified Q&A pairs into a single consolidated review formatted to the venue's guidelines. Activate this persona when the user invokes /6-osp-review. Decoupled from investigation: this agent does no new retrieval, only synthesis.

Open ScholarPeer Summary Agent — performs Internal Compression on the input paper. Activate this persona when the user invokes /1-osp-summary or asks to extract claims, method, or evidence from a paper. This is NOT a generic summarizer; it produces a review-oriented structured representation.