open-scholar-peer: Instructions file for Kiro

Open ScholarPeer — Always-On Rules

These rules apply automatically in any project where Open ScholarPeer is installed.

Brain protocol (apply on every invocation)

1. Read .brain/session.json first to understand current state.
2. Load only the artifacts in the active step's reads: contract. Do not load the full .brain/ directory. A listed artifact that is missing is an input you do not have — not a reason to stop.
3. Update session.json around every step. Set started_at when you begin. On finishing, set the matching phases.<name> block to completed, set completed_at, and update resume_from.
4. When the user moves past a step without running it, record that. Set that phase's status to "skipped" and its skip_reason to what they told you, or to "user moved on" if they said nothing. Do this the moment you start the later phase — a skip nobody wrote down is a silent degradation, and MANIFESTO rule 8 forbids those. The four permitted values are pending, in_progress, completed and skipped.
5. Carry every skip forward. Name it in the artifact's ## Provenance and say what it cost this phase — not what it cost an earlier one, which has already been said where it belonged. Each phase reports its own consequence, once. /6-osp-review collects them all under ## What this review did not have, so whoever reads the review can tell a thin corpus from a thorough one. State it; do not repeat it, and never dress it as a warning.
6. Re-runs overwrite with a warning. If a step is already completed, print one warning, then proceed.

Persona discipline

• Each numbered command (/N-osp-*) activates exactly one persona skill (osp-<name>-agent). Do not blend personas.
• The Q&A engine (/5-osp-qa) uses two personas at once: Query Agent (main thread) and Answer Generator (subagent, or self-reflection on tools that lack subagents).
• The orchestrator (osp-orchestrator) never performs review work — it only routes.