safer-docs-reader

Automate GitHub PR review by spawning an ephemeral team of four haiku sub-agents that read the artifact cold. The orchestrator collects their per-persona judgments, aggregates them via severity-weighted consensus, and posts the final SHIP or REVISE verdict back to the PR thread.

Multi-agent documentation auditing is a process where multiple haiku sub-agents evaluate a docs artifact cold. This approach prevents context leakage and ensures consistent review quality by aggregating individual structured verdicts through a severity-weighted consensus mechanism.

You can review local markdown files by resolving the file into a deterministic input for evaluation. The workflow evaluates the content across three rounds and publishes the final aggregated SHIP or REVISE report directly to standard output.

Severity-weighted consensus for AI issue review works by collecting structured verdicts from four haiku sub-agents. The orchestrator weighs the severity of each identified issue to decide a final SHIP or REVISE outcome, ensuring consistent review quality without context leakage.

You need gh CLI authentication to run automated AI docs review on GitHub artifacts. The workflow requires gh CLI access to resolve issue and PR bodies, alongside ephemeral TeamCreate and Agent tools to spawn the sub-agents.

The limitation of this multi-agent artifact review is a strict round limit of three rounds with explicit gating for round three. This constraint ensures deterministic evaluation but prevents extended iterative refinement beyond the third round.