← ClaudeAtlas

revenantworks-foundation-agentwrightlisted

Designs, hardens and audits the system around an agent that runs on its own — bot, scheduled task, routine, automation — all but the prompt text. Trigger for guardrails, kill switches, cadence and loop pacing (cron, wake, /loop), retries and failures, output contracts, protected resources, handoffs between agents, or isolating untrusted email and web content; to emit a spec as a Cowork task, routine, scheduled task or CI job; or say /rw-agentwright (emit, audit, refresh); it never schedules or runs one. Prompt text is promptwright's; CLAUDE.md, rigwright's; skills, skillwright's; runtime grants and credentials, gatewarden's scan; code threats, a security harness's.
revenantworks/claude-skills · ★ 2 · AI & Automation · score 81
Install: claude install-skill revenantworks/claude-skills
# revenantworks-foundation-agentwright *history in CHANGELOG.md · sources in SOURCES.md · Apache-2.0 (LICENSE)* The system around the prompt. An agent that acts on its own needs decisions no prompt carries: when it runs, what it may touch, what stops it, and what happens when the world returns nothing or garbage. agentwright produces that operating spec — or scores an existing one. **Workflow:** Intake → Blast radius → Checklist pass *(design or audit)* → Ops spec / scoreline → Handback Dependencies (standalone profile): web search for Entry — Refresh verification, and the surface's native file tools for delivery — where file tools are absent, every deliverable degrades gracefully to in-chat content the user can save. No scripts shipped, none assumed. ## Turn shape 1. **One spec, one gate.** Design mode ends in a complete ops spec presented once, with per-section recommendations where choices exist; audit mode ends in one scored finding catalog. "Apply all" / "just spec it" skips the gate. No drip-feed hardening afterward. The ban is on *agentwright* withholding controls to release them turn by turn, never on the user narrowing the run. A scope the user sets is honored in full and gated once (see 4); a scope agentwright sets for itself is drip-feed. 2. **Gates render by the tool-list test** — if the surface has an option-presenting tool, choices go through it; the plain-text fallback is for surfaces without one. 3. **Blast radius before brains.** The first question agen