Run an opt-in deep, single-agent engineering review with risk analysis, bounded capture, verification claims, and a machine summary. Use when the user explicitly requests this Skill, a machine review bundle, or strict evidence for a high-risk/release change. Do not invoke automatically for routine repository work merely because Rootloom is installed.
Explicitly create, refresh, and validate a concise evidence-backed AGENTS.md managed block for a Git repository or a genuine nested module boundary. Use only when the user asks to persist project guidance. The SessionStart Hook provides temporary read-only context and never invokes this writer. Preserve unmarked existing guidance and route semantic polishing to refine-project-guidance.
Plan and execute high-risk engineering changes involving public APIs, schemas, migrations, persisted data, security, architecture, production dependencies, infrastructure, release, deployment, or destructive effects. Requires impact analysis, compatibility, rollout, rollback, and explicit authorization for external actions; plans stay task-local unless durable planning is explicitly requested.
Experimentally initialize, retrieve, and explicitly update Rootloom's repository-owned project memory for architecture, known risks, durable decisions, and evidence-backed failure lessons. Use only when explicitly requested and durable history will change a future engineering decision. Memory is advisory and outside stable Core.
Plan, install, inspect, update, or roll back Rootloom Personal Core in a user's Codex home. Supports Skills-only, guidance, and the recommended personal preset. Use when the user explicitly asks to install, configure, bootstrap, repair, audit, update, reduce, or remove Rootloom. Never overwrite user-owned files without showing the plan and obtaining exact replacement authorization.
Perform a review-only analysis of code, diffs, pull requests, migrations, or architecture changes. Report evidence-backed findings first by severity with exact locations and concrete fixes; do not modify code unless the user explicitly asks for fixes.
Implement or fix repository code with root-cause diagnosis, focused scope, working-tree protection, and proportional verification. Use for bugs, features, refactors, tests, and ordinary multi-file changes; do not use for review-only work or high-risk contract/migration/release changes.
Create or update a durable repository-owned engineering decision record with alternatives, evidence provenance, consequences, verification, and supersession links. Use when architecture ownership, public or persisted contracts, dependencies, security posture, data strategy, or operational policy changes in a way future agents and maintainers must remember; do not use for routine bug fixes, mechanical edits, transient incidents, or plans that have not been accepted.