Evidence Synthesiser — Development agent for Claude Code
Makes the synthesis pass over an evidence subgraph once its papers have been extracted: merges ids that parallel extractors minted twice, draws the links that span papers, and checks that the evidence.
How to install Evidence Synthesiser
Installs to ~/.claude/agents/phelps-sg-scibraid-evidence-synthesiser.md
mkdir -p ~/.claude/agents && curl -fsSL https://raw.githubusercontent.com/phelps-sg/scibraid/HEAD/agents/evidence-synthesiser.md -o ~/.claude/agents/phelps-sg-scibraid-evidence-synthesiser.md Restart Claude Code, or start a new session, for it to be picked up.
What Evidence Synthesiser does
name: evidence-synthesiser description: Makes the synthesis pass over an evidence subgraph once its papers have been extracted: merges ids that parallel extractors minted twice, draws the links that span papers, and checks that the evidence is connected to the hypotheses in contention. Use from the evidence-subgraph skill after extraction. Give it the subgraph slug and the extractors' reports. model: fable tools: Bash, Read, Write, Edit, Grep, Glob, WebFetch skills:
- evidence-subgraph
Alternatives in Development
- People Profiler — Bulk create or update person notes from Slack profiles 4.6k ★
- Reverse Inventory — Pour un projet legacy déposé dans workspace/old/{LegacyProject}/, cartographie déterministe : détecte langages 193 ★
- CI Watcher — Watch PR checks for the current branch and report pass/fail with relevant links and concise next steps 144 ★
Full documentation available on GitHub
View Source RepositoryRelated Agents
Checklist Deduper
Merges multiple batches of checklist items from checklist-generator into a single deduped checklist. Preserves
Synthesis Planner
Plans the structure and narrative arc for focused, insight-driven literature reviews. Designs tight outlines (
Proteus Guide
Proteus review guide. After a milestone merges, opens the review issue with the human's brief and evidence lin
Implement Manager
WF-MAX Manager for W2 implementation wave. Spawns 5-7 implementers (one file_claim each), merges results, repo
Code Filter
Independent falsify pass for review-gate findings. Receives unified diffs + a findings list and returns the ID
Parallel Search
Low-level paired-search primitive. Given ONE query, reads Google through formal LCU MCP and runs WebSearch wit
Related Skills
Audit Strict
Multi-pass consensus audit — runs the audit twice with different prompts, only reports consensus findings. Agg
Paper Read
Read one or more papers from the knowledge garden by their IDs and answer a question.
AI Work Log
Work with the AI Work Log paper logbook (one spread per AI chat or project, spread ids S01-S60, four index-car