← ClaudeAtlas

independent-reviewlisted

Spawn one or more uncontaminated subagents to review completed work, then a separate uncontaminated verifier subagent to validate their findings against the work itself. Use for any produced artifact: code, text, specs, emails, drafts, documents. Called with /independent-review
renatodvc/rstack · ★ 1 · Code & Development · score 74
Install: claude install-skill renatodvc/rstack
Review completed work using fresh subagents that have not seen the conversation, then verify their findings with another fresh subagent. Fixing issues is out of scope: this skill ends by presenting a report. ## Contamination rules (apply to every subagent spawned by this skill) - Spawn with no conversation context: use a standard subagent, which starts fresh. Spawning as a fork (subagent_type: "fork" or context: fork) is forbidden, regardless of session defaults, because forks inherit the full conversation. - Provide only: the goal of the work, a factual description of what was done, the work itself, and any decisions the user explicitly deferred or exceptions the user explicitly granted. - Never share your own reasoning, opinions, doubts, or expectations about the work. - When the work lives in files, pass file paths and let the subagent read them. When it exists only in the conversation (an email draft, a plan, a snippet), pass the verbatim content, never a paraphrase. ## Step 1: Scope Identify the artifact under review, the original goal and requirements, and any user-deferred decisions or granted exceptions. Decide the number of reviewers based on size and stakes: - Small or low-stakes work: 1 generalist reviewer. - Larger or higher-stakes work: 2 or 3 reviewers, each with a distinct lens. Choose lenses that fit the artifact, for example: - Correctness: is the content right, sound, and internally consistent? - Completeness: does it fulfill every stated requiremen