private-transfers-benchmarks

Per-property guidance for the Cryptography property group. Invoke when evaluating, reviewing, or editing notes/values for Verifiability, Post-quantum secure, Number of secrets, or Client-side proving. Cross-cutting rules in scripts/research-prompts.ts still apply on top — particularly WRITING_RULES

Per-property guidance for the Trust / robustness property group (drawn from the schema's Decentralization & Security and Verifiable groups, excluding maturity which lives in property-rules-timing). Invoke when evaluating, reviewing, or editing notes/values for Censorship resistance, External network dependence, Escape hatch, Open source, Upgradeability, or Third-party inspectability. Cross-cutting rules in scripts/research-prompts.ts still apply on top.

End-to-end evaluation pipeline for a single protocol — discover sources, run the citations research script, review the output against rules, run an independent second-pass review, and report, applying the qualifying rule-gap improvements to the rule files and leaving them uncommitted for human review.

Per-property guidance for timing-related properties — Time-to-finality, Deposit time, Withdraw time, and Implementation maturity. Invoke when evaluating, reviewing, or editing notes/values for any of these. Cross-cutting rules in scripts/research-prompts.ts still apply on top.

Per-property guidance for the State property group. Invoke when evaluating, reviewing, or editing notes/values for Private state model, Private Data Storage, Client-side indexing, or Private State Scalability. Cross-cutting rules in scripts/research-prompts.ts still apply on top.

Find authoritative source URLs for each property of a protocol evaluation using WebSearch/WebFetch, then write the results to scripts/research-cache/{id}.json so the API research script can consume them without spending API credits on web search.

Per-property guidance for the Privacy property group. Invoke when evaluating, reviewing, or editing notes/values for any of Anonymity, Confidentiality, Asset privacy, or Plausible deniability. Loads only the privacy-group rules so the active context stays focused on the property at hand. Cross-cutting rules in scripts/research-prompts.ts (WRITING, CROSS_CHECK, VALUE_FORMAT, SOURCE_SELECTION, REVIEW_ONLY) still apply on top.

Independent second-pass review of a protocol evaluation. Run AFTER /review-evaluation has produced its first report. This skill spawns a fresh subagent with no exposure to the first review's findings, has it review the same evaluation from scratch against the same rules, and diffs the two reports so anchoring bias from the first pass is surfaced. Use when an evaluation needs higher-confidence sign-off, especially before merging or shipping.

Per-property guidance for the Composability property group. Invoke when evaluating, reviewing, or editing notes/values for Access to DeFi or Programmability / Generality. Cross-cutting rules in scripts/research-prompts.ts still apply on top.

Devil's-advocate review pass over a protocol evaluation. For each property, a fresh sub-agent argues the closest counter-value to surface anchoring failures the same-session reviewer might have missed. Runs as Phase C.5 of /evaluate-protocol, between the first-pass review (Phase C) and the auto-patch phase (Phase D).

Per-property guidance for the Compliance property group. Invoke when evaluating, reviewing, or editing notes/values for Type of compliance, Layer of enforcement, Enforcement entities, Point of enforcement, Selective disclosure: viewing entity, or Selective disclosure: viewing control. Loads only compliance-group rules so the active context stays focused. The four-property compliance quartet must be internally consistent (see CROSS_CHECK_RULES in scripts/research-prompts.ts) — that cross-cutting rule is the most important constraint for this group.

Review a protocol evaluation for accuracy, consistency, and rule compliance. Use when asked to review an evaluation.