Every skill is checked before it is listed.

CRE skills run against confidential files: rent rolls, T-12s, IC memos, tenant PII, sponsor financials under NDA. A bad skill could leak that data, reach for an undisclosed credential, or hand you a number you shouldn't trust. So the rubric weights what a skill is allowed to do (its purpose, its instruction scope, how it installs) above whether it writes a file. Vetting a skill before you trust it with a deal is the same discipline as vetting a vendor before granting data-room access.

Weights reflect how much damage a skill can do: its purpose, instruction scope, and how it installs matter more than whether it writes a file. Scores are 1–5 per dimension; the weighted total is normalized to a 0–5 scale (weights sum to 11). Every skill on the site is reviewed against this rubric by the maintainer, which reports a concern band. The Verified/Caution/Flagged verdict tiers (≥ 4.0, 3.0–3.99, below 3.0) are a separate, stronger claim reserved for a formal third-party audit, which none has yet.

These verdict tiers apply only when a real, committed audit record exists. No skill is audit-backed today, so no skill carries a Verified, Caution, or Flagged verdict. The owner-reviewed methodology checks below report a concern band instead.

These figures describe the catalog, the rubric, and the owner-reviewed methodology checks. They are not formal audit results: every skill carries a maintainer review with a concern band and score, but zero skills have been formally or independently audited.

How the 124 owner-reviewed skills fall across the concern bands. Most are focused, local read-and-reason tools and land in Low concern; the workspace routers that ship their own runtime subsystem are the Moderate outliers. Elevated concern is reserved for a concrete red flag — meaningful runtime complexity, credential handling, an external connector, or mutation behavior — and none of these read-and-reason skills carry one, so the band is empty. The bands are what the maintainer review found against the rubric, not a rubber stamp:

This is a real, committed methodology check, the same maintainer review shown on the skill's own page, not a hand-picked illustration. The site owner reviewed the skill against the rubric: its catalog entry, manifest, declared runtime behavior, and any calculator files. It is a maintainer review, not a formal or third-party audit.

• Each methodology check is a maintainer review: the site owner scores the skill against this rubric from its declared catalog/manifest metadata, current plugin source, and any calculator or runtime files. It reviews the skill's security surface, not the correctness of its analysis or its full runtime behavior under every input.
• The plugin declares no tool/capability permissions (no allowed-tools), so the review cannot assert what a skill reads, writes, or accesses at runtime. Dimensions that depend on declared capability are scored conservatively, and effective capability is set by the agent you run the skill in.
• This is a maintainer review, not a formal or independent third-party audit. Zero skills are audit-backed. A low concern band means the review found no concrete red flag against the rubric; it is not a certification of safety.
• The plugin is open source and changes independently; an upstream commit can alter a skill's behavior after any check here, so a check reflects a point in time. If the upstream version drifts from what the site shows, the check auto-hides rather than mislabel a changed skill.
• Skill behavior depends on the host agent, the model version, your inputs, and your local environment; identical instructions can produce different results across setups.
• Everything here is provided "as is," without warranty of any kind, under the Apache License 2.0.
• You are solely responsible for any sensitive, confidential, or regulated client and portfolio data you choose to put in front of any skill or agent.
• Nothing on this site is investment, legal, tax, or accounting advice.

Every text color pairing used in this site is verified against the WCAG 2.1 AA threshold at build time. The table below is generated from the same source module the test suite asserts, so it cannot drift from the actual tokens in use.

A skill’s trust signal can only ever be one of these four states. The site sits at “methodology assessed” today: every skill is checked, no skill is formally audited, and the two can never be confused because they never share a render path.

1. Not assessed

   The baseline. A skill with no committed assessment pinned to its current version shows “Not formally audited” and no score. For example, a skill whose upstream version has drifted from what this site shows, so its methodology check no longer matches.

2. Today: methodology assessed

   Every skill carries a maintainer review that scores it against the rubric above (its catalog entry, manifest, declared runtime behavior, and any calculator files) and reports a Low/Moderate/Elevated concern band. This is the site owner's review, not a formal or independent third-party audit: it can surface a concern but never certifies a skill as safe, and it never uses the verdict words below. Each skill still shows “Not formally audited,” because a maintainer review is not an audit.

3. Next: audit pending

   When a skill is queued for a real, evidence-backed review, the contract supports an “audit pending” state. It shows no formal score or verdict (only that a review is in progress), so a pending review can never be mistaken for a completed one.

4. Audit-backed

   A skill earns a formal Verified/Caution/Flagged verdict only when a committed audit record exists that pins to the exact reviewed commit and version, scores all five dimensions, and whose verdict is derived from that score. If the upstream plugin changes, the record stops matching and the skill returns to neutral. A trust signal can never outlive the code it described.