agentsclimarketplace
Sign in

Reviewing decisions

Skill vemodalen-x/VEMO_SKILLS/skills/research/reviewing-decisions

Reliability ruleset for technical design decisions — a lightweight design-review discipline. Each key decision must carry 6 fields (options/evidence/trade-offs/assumptions/failure-modes/validation); a cross-model red-team pass critiques the doc before publish. Use when the research-solution agent finalizes the solution_document. Owned by Framework 1 (research). Generic; reviewer models/decisions are instance values.From its SKILL.md

Install
npx -y skills add vemodalen-x/VEMO_SKILLS --skill reviewing-decisions

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.0 KB, 632 tokens by cl100k_base, as published. Nobody here has run it

Decision Review (reliability discipline)

Makes design decisions reliable, not just stated. Two parts: a per-decision 6-field floor (checkable), and a cross-model red-team critique before publish. Self-built, borrowing the ideas of structured-MADR (machine-checkable decision fields) and adversarial-review (cross-model critique) — see the research catalog borrow-note. Pairs with structuring-solution-docs (its S8/MADR blocks ARE these decisions). Generic (skill_spec §9).

When to use

  • The research-solution agent finalizes the solution_document (Wildtarsier §1.9); apply before 准出 (Gate 2).

A. Per-decision 6-field floor (准出 completeness — missing any = not done)

Every key design decision MUST carry all six. The 准出 check (Gate 2, compliance) verifies presence — this is a completeness criterion of the solution_document output, NOT a new gate (the loop stays the existing two gates):

  1. Options compared — ≥2 candidates with pros + cons each (no single-option "decision").
  2. Evidence with provenance — each claim sourced + tagged 实测 / 文献 / 推测 (defer to technical-report-style R4).
  3. Trade-offs — what the chosen option costs, stated plainly (no one-sided upside).
  4. Assumptions — what must hold for the decision to be valid (so they can be checked later).
  5. Failure modes — how it can go wrong + the mitigation or detection.
  6. Validation — how the decision will be verified (experiment / metric / PoC) — ties to the feasibility region.

B. Cross-model red-team (pre-publish adversarial pass)

  • Before the solution_document is published, a different agent / model reviews it adversarially: actively look for weak evidence, unstated assumptions, missing failure modes, one-sided trade-offs, and over-claims.
  • Findings are recorded; material ones are resolved (or explicitly accepted with reason) before publish.
  • This is a step inside the existing 准出, not a third blocking gate — it runs out-of-band like the governance behavior-audit, and its output is advisory input to the compliance Gate-2 decision, not a separate gate.

Rules

  • A decision missing any of the 6 fields (§A) → the solution_document is incomplete → 准出 (Gate 2) fails.
  • The red-team pass (§B) must have run before publish; its findings + resolutions are recorded in the task ledger.
  • Decoupling (skill_spec §9): generic. Which model(s) red-team and which decisions are "key" are instance/project values — e.g. project_profile.yaml may name the red-team reviewer model; not hardcoded here.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Gives 0 of the 12 instructions most review quality skills give in 632 tokens

Counted across 1,048 of the 1,783 authors here whose files we hold, read 2026-08-07

  • Ask questions one at a timein 81 of 1048, across 64 files
  • Provide a recommended answer for each questionin 73 of 1048, across 50 files
  • Explore the codebase instead of asking answerable questionsin 66 of 1048, across 42 files
  • Resolve dependencies between decisions one-by-onein 42 of 1048, across 17 files
  • Interview the user relentlessly about the planin 38 of 1048, across 13 files
  • Order findings by severityin 31 of 1048
  • Resolve each branch of the decision treein 27 of 1048, across 5 files
  • Run a grilling sessionin 26 of 1048, across 5 files
  • Update CONTEXT.md immediately when a term is resolvedin 26 of 1048, across 11 files
  • Propose precise canonical terms for vague languagein 25 of 1048, across 7 files
  • Create documentation files lazilyin 24 of 1048, across 5 files
  • Assign severity to every findingin 24 of 1048

Said here and by no other author read

  • include all six fields per decision
  • state trade-offs of the chosen option
  • list assumptions for the decision
  • describe failure modes and mitigations
  • specify how to validate the decision
  • run a cross-model red-team pass before publish

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.