research-reviewlisted

# /research-review — parallel persona audit This skill runs four narrow review personas against a target file and merges their findings. Each persona returns structured JSON with severity + file:line + recommendation. The orchestrator dedupes overlapping findings and presents a single punch list. ## When to invoke - **Before shipping** a new `continuous-research/findings/**` file or synthesis doc - **Before publishing** an article that depends on research claims - **After a large research round** where multiple files were added or modified - **During `/refresh` audits** when compounded entries point at research files as stale ## What it does NOT do - Edit the file. The orchestrator presents findings; Antti (or a follow-up skill) decides and applies fixes. - Propose new sources. Audit only. - Run continuously. On-demand. ## The flow ### Step 1 — Target resolution Accept `$ARGUMENTS[0]` as the file path. If missing, ask for one. Do not run on directories — one file per invocation. ### Step 2 — Parallel launch Launch all four persona agents in a single message (parallel tool calls). Each gets: - The target file path - A copy of `continuous-research/research-rules.md` prepended to the prompt - Explicit instruction: return JSON per the agent's output format, no prose Agents: - `source-type-auditor` — flags unlabeled URLs and Level 0 sources used as evidence - `zombie-stat-detector` — flags round-number AI stats and known zombies - `freshness-checker` — flags sources o