@ephremyuan/citeagent

OpenCode plugin for CiteAgent — AI research knowledge infrastructure with Merkle-verified retrieval, citation-indexed search, and trace-bound chat.

Install

bunx @ephremyuan/citeagent@latest install

The installer automatically:

• Adds the plugin to ~/.config/opencode/opencode.jsonc
• Deploys skills to ~/.config/opencode/skills/
• Deploys agent configs to ~/.config/opencode/agents/
• Deploys rules to ~/.config/opencode/rules/
• Generates agent model mappings in ~/.config/opencode/citeagent.json

Options

| Flag | Description | |------|-------------| | --reset | Overwrite existing configuration | | --dry-run | Simulate install without writing files |

Prerequisites

# Optional: citeindex CLI for document ingestion (PDF, URL, media)
# If not installed, cite_ingest/cite_tantivy_index will warn but other tools work fine
uv tool install citeindex

# Optional: OCR support
sudo apt install tesseract-ocr

# Optional: LLM backend (for chat/generation)
# https://ollama.ai

No Python runtime required — all citation tools run natively in TypeScript.

Uninstall

1. Remove "@ephremyuan/citeagent" from ~/.config/opencode/opencode.jsonc plugin array
2. Remove config: rm ~/.config/opencode/citeagent.json
3. Remove assets: rm ~/.config/opencode/skills/citeagent-*.md ~/.config/opencode/agents/citeagent-*.md ~/.config/opencode/rules/citeagent-*.md

Agents

| Agent | Mode | Description | |-------|------|-------------| | citeagent-researcher | primary | Academic research with citation-verified evidence | | citeagent-verifier | subagent (hidden) | Independent Merkle proof audit | | citeagent-explore-corpus | subagent (hidden) | Fast corpus search and browsing | | citeagent-ingestor | subagent (hidden) | Document ingestion (PDF, URL, media) | | citeagent-reviewer | subagent (hidden) | Systematic literature review |

Tools

The plugin provides 25+ tools via the native TypeScript CiteAgentEngine:

• cite_search — BM25 full-text search
• cite_verify — Merkle proof verification
• cite_render — CSL citation rendering (Chicago, APA, MLA...)
• cite_ingest — Document ingestion with Merkle hashing
• cite_tree / cite_tree_traverse — PageIndex document tree
• cite_argument_query — Argument graph (claims, contradictions)
• cite_memory_* — 4-tier persistent memory (working → episodic → long_term → corpus)
• And more (see source)

Architecture

User question
     │
     ▼
citeagent-researcher (OpenCode agent)
     │
     ├── cite_search ──→ CiteAgentEngine (TypeScript, in-process)
     ├── cite_verify ──→ CiteAgentEngine
     ├── cite_ingest ──→ citeindex CLI (optional sidecar)
     │
     ├── @citeagent-explore-corpus (OpenCode subagent)
     └── @citeagent-verifier (OpenCode subagent)

All tools except cite_ingest/cite_tantivy_index run natively in TypeScript — no Python subprocess or MCP stdio bridge required. The engine reads the on-disk corpus directly and implements BM25 search (MiniSearch), Merkle verification, CSL rendering, memory store, and audit trail in-process.

License

MIT