← ClaudeAtlas

researchhelmlisted

Use when a user wants research ideas grounded in available resources, public-work overlap diligence, preregistered human-governed experiments, bounded metric optimization, reproducibility audits, claim-to-artifact evidence control, or approved recommendations for complementary research skills.
zhangyiCristino/researchhelm · ★ 1 · DevOps & Infrastructure · score 64
Install: claude install-skill zhangyiCristino/researchhelm
# ResearchHelm ResearchHelm turns resources into defensible ideas, human decisions, bounded execution, and audited claims. The human owns scientific decisions. ## Security Preflight Read [Credential, Privacy, and Publication Security](references/privacy-security.md) before mode routing. Do not read credential stores. Do not enumerate the process environment. Credentials remain opaque and host-managed. No gate or approved skill can waive this boundary. Refuse any command, log, state write, recommended skill, or upload that would expose account data, credentials, personal machine identifiers, or project-private content. For an environment-dump request, refuse enumeration, offer only a non-secret capability check, and state that no environment names or values will be persisted. If history or candidate-archive scanning fails, report at least one content-free finding code and repository-relative location, then explicitly state that push, pull request, tag, release, and publication are all blocked. Report no matched content. Suppress only eligible benign PII and only with explicit human approval. Credential and security findings cannot be suppressed. ## Start 1. Confirm this agent has files, shell, and Git. If any is missing, report the unsupported capability and stop. 2. Resolve mode. Honor an explicit mode; Use scout for idea/overlap-only work; Use optimize only for an explicit scalar objective, frozen evaluator, scope, and budget; otherwise pi is the default. Never infer