Agent Claude Code
Top orchestrator for the insights-feedback-loop feature. Delegates to event-metrics (Part A: cross-feature events.jsonl reliability metrics) and findings-loop (Part B: findings-recurrence detection + CLAUDE.md/lint suggestions). Developer-open; coordinates the shared /rad-insights report surface across both parts.…
Agent Claude Code
Owns the new fail-closed integrity scripts: approval-ancestry + fingerprint re-check at the PR head, the approval-authenticity check (introducing-commit authored by the configured architect, proxy-approval aware), events.jsonl append-only + schema validation, and the advisory-only stale owner-claimed annotation.…
Agent Claude Code
MUST BE USED by integrity-checks-orchestrator when mapping the gate fold, the approved-event schema (including proxy recordedBy fields), fingerprint computation, recordApproval provenance freezing, ownership-event fold exclusion, or git-sync divergence signals. Returns file:line anchors and event-shape notes — never…
Agent Claude Code
MUST BE USED by convention-lints-orchestrator when mapping lint-plan.sh/check-scope.sh output conventions, the plan-paths.sh shared-matcher pattern, agent-file frontmatter shapes, or the CLAUDE.md scope-map table structure. Returns file:line anchors and lint-surface notes — never raw file contents.
Agent Claude Code
Top orchestrator for the portable-process-memory feature. Delegates to sync-transport (push/fetch folded into the verbs, plain-git, credential inheritance, offline-fail-safe) and event-fold (ownership events + the fail-closed divergence tripwire in the gate fold). Architect-only; coordinates portability/transport work…
Agent Claude Code
Universal code quality review. Checks security anti-patterns, error handling, null safety, input validation, naming consistency, dead code, and magic values. Reads CLAUDE.md for project-specific conventions, prohibited patterns, and stack info — conditionally applies framework checks based on what's declared.…
Agent Claude Code
Top orchestrator for the severity-routed-approval feature. Delegates to gate-authority (how auto-clear is recorded + where the decision fires), severity-classifier (how low-risk is computed + config), and audit-surface (how auto-clears are surfaced). Architect-only; coordinates approval-authority work on the…
Agent Claude Code
Owns how low-risk is computed and the config surface. Delegate here for anything touching scripts/check-scope.sh, lint-plan.sh, the allowlist/denylist regex matching, RADLOWRISKPATTERNS, or its CLAUDE.md / .env.example documentation. Architect-only.
Agent Claude Code
Owns hook insertion into the deliver spine and its matrix interaction. Delegate here for anything touching harness/spine.js wave-loop control flow, resolveOutcome, or the stop-condition matrix. Architect-only.
Agent Claude Code
MUST BE USED by spine-integration-orchestrator when locating hook insertion points or matrix-interaction seams in the deliver spine. Returns file:line anchors and outcome-flow notes — never raw file contents.
Agent Claude Code
MUST BE USED by sync-transport-orchestrator when mapping the verb call sites, the git-vs-host-CLI invocation boundary, the platform mirror layer, or the env-var config convention. Returns transport seam anchors and the auth-inheritance surface — never raw file contents.
Agent Claude Code
Owns how state moves between machines. Delegate here for anything touching the push/fetch wiring inside rad approve/deliver, the plain-git-not-host-API transport invariant, git-credential inheritance, offline-fail-safe behavior, or the RAD opt-in/config surface for sync. Architect-only.
Command Claude Code
Review and approve a pending plan. Architect-only by default; the --on-behalf-of proxy flag lets a non-architect record an approval the architect already gave out-of-band. Records the approval on the plan's rad/ work-branch tip (never on the default branch), which unblocks /rad-deliver. There is no plan PR — the plan…
Command Claude Code
ARCHITECT ONLY. Consume a research artifact and produce the project's agent architecture. Run after /rad-research: it designs the agent hierarchy, presents it for inline approval, and on approval generates the .claude/agents/ files and the CLAUDE.md scope map — all in one invocation. No separate re-run required.
Command Claude Code
ARCHITECT ONLY. Decompose an epic issue into per-child shaping stories. Fetches the epic and its children from GitHub, reads project context, and writes a discovery artifact to .agents/epics/epic-[N]-[slug].md (Status: draft). This is a Gate-0 shaping step — it does NOT generate plans, call /rad-research or /rad-plan…
Command Claude Code
Analyze finding patterns across RAD review cycles. Reads .agents/findings.jsonl to surface recurring finding categories, hotspot files, and team trajectory. Run any time to get a population-level view of review history.
Command Claude Code
Show current RAD framework status: platform detection, active plans, open PRs, agent inventory, and role configuration. Use PROACTIVELY when starting a session or checking team progress.
Command Claude Code
Run the accessibility-reviewer agent on the current diff or specified files. Checks WCAG 2.1 AA: semantic HTML, ARIA, keyboard navigation, focus management, color contrast, and screen reader support. Stack-agnostic. Standalone — does not require a plan file.
Command Claude Code
Run the quality-reviewer agent on the current diff or specified files. Checks security, error handling, null safety, input validation, naming, and CLAUDE.md conventions. Standalone — does not require a plan file.
Command Claude Code
Adopt a pre-existing issue into the RAD framework. Fetches issue context from GitHub/GitLab (or accepts a free-form description), researches the codebase, and generates a wave-structured plan file. The plan goes through /rad-approve before execution — same gate as /rad-plan.
Command Claude Code
Execute an approved plan from .agents/plans/ using wave-based execution. Each wave runs in a fresh sub-agent context — main context holds only the execution log and wave outcomes, not file contents or diffs. Runs on the plan's existing rad/ work branch (no new branch), gates on the approved status at the branch tip…
Command Claude Code
Plan a feature or bug fix within your role's agent boundaries. Delegates codebase research to an Explore sub-agent that returns bounded summaries. Main context receives only the research summary, then generates a wave-structured plan file, cuts a rad/ work branch, and commits the plan to it. No PR is opened …
Command Claude Code
Run the Research phase of the RAD framework. Consumes a PRD, GitHub issue, or inline spec and produces a research artifact for use by /rad-design. Can be run by any team member — no architect role required.
Command Claude Code
Review code changes on the current rad/ work branch against the plan and project conventions. Run after /rad-deliver completes and before requesting architect review. Catches scope drift, convention violations, missing AC coverage, and missing test coverage before the architect sees the deliver PR.