| name | classical-lexicon |
| description | Ground CoSAI Risk Map terminology in established security terms of art. Use when authoring or critiquing a Control, Risk, Component, or Persona title or description to check a proposed term against the canonical (NIST-first) vocabulary, replace an invented term with its established equivalent, generalize a product/protocol-specific name to its role, or confirm a term is already grounded. |
Classical Lexicon
Check Risk Map terminology against established security terms of art. The governing rule: never invent a term when an established term of art already exists. This skill grounds a proposed term, regrounds an invented one, or confirms an existing term is already sound — it does not rewrite content wholesale.
When to use
- Authoring a new Control / Risk / Component / Persona title or description.
- Critiquing a draft for coined or product-specific terminology.
- Deciding what to call a capability, role, mechanism, or locus — including a persona/role name, not only a control or risk mechanism term.
When not to use
- Framework mapping IDs (MITRE ATLAS, NIST AI RMF, etc.) — that is the framework-mappings audit tooling's job.
- Prose style/format rules (bold, sentinels, length) — that is the prose-subset linter's job.
Procedure
-
Extract candidate terms from the input: the nouns and noun-phrases that name a capability, security measure, role, mechanism, or architectural locus. These are where terminology matters.
-
Look each up in the lexicon: references/lexicon.md.
-
Classify each candidate:
- Grounded — already an established term of art. Accept; name the source.
- Reground — an invented or informal term with an established equivalent. Return the canonical term and the mapping.
- Generalize — the term encodes a specific product or protocol (e.g. an
MCP-prefixed name). The identity is the role/locus; the product is an attribute. Return the role-grain term; keep the product only as an example.
- Novel — flag — no established term of art found. Do not coin one silently (see the maintainer-flag vocabulary below). "novel-flag" applies only when no established term of art exists in the field, not merely when the term is absent from this pinned lexicon snapshot; when you can ground a term via general NIST-first sourcing knowledge that the snapshot lacks, ground it and propose the lexicon addition rather than flagging it novel.
3b. Record the international/standards equivalent — unconditionally, for grounding verdicts only. This obligation applies to grounded and reground candidates — verdicts that ground a term in standards. For every such candidate, record its international/standards equivalent(s) (ISO/IEC, EU AI Act, ENISA, XACML, or other recognized non-US/international body). Do this every time a term is grounded or regrounded — not only when NIST is silent or the term is contested. A maintainer flag fires only when this equivalence check turns up a problem: the equivalent is missing (
D3b-parochialism) or the equivalent conflicts with the NIST-first term (naming-conflict or substantive-conflict, see below). When the equivalent is present and agrees, record it and raise no maintainer flag.
A generalize verdict is a naming-altitude move (product/protocol name → role/locus identity), not a grounding-in-standards move. It carries no international-equivalence obligation: record the international equivalent field as "not applicable" and raise no equivalence flag (D3b-parochialism, naming-conflict, or substantive-conflict) on a generalize verdict.
-
Record the counterfactual for every reground/generalize/novel result: proposed → chosen → source (or "none found"). This is inspectable reasoning, not a silent rename.
NIST-first priority (ADR-031 D3a)
NIST's term is the term unless there is a strong, documented argument that it fails. Prefer, in order: NIST (SP 800-53 / 800-207 / 800-162, AI RMF), then other US/international standards bodies and classical literature (RFC, OWASP, MITRE, ISO, foundational papers).
If a contributor argues the NIST term fails and a new term is needed, do not decide that here. Surface it as a documented deviation for the maintainer to take into CoSAI governance review (issue / discussion). This skill grounds and flags; maintainers adjudicate.
Global equivalence, unconditionally recorded (ADR-031 D3b)
A NIST label is adopted only after the concept is corroborated as globally recognized. For every grounded or regrounded term — not only contested or NIST-silent ones — record the cross-standard equivalence set: the canonical (NIST-first) label alongside its equivalent(s) in recognized international standards (ISO/IEC, EU AI Act, ENISA, XACML, or other national/international bodies). This is unconditional within its scope: run the equivalence check on every grounded/regrounded term, every time.
This equivalence check applies only to grounded and reground verdicts — the verdicts that ground a term in standards. A generalize verdict is a naming-altitude move, not a grounding move, and carries no equivalence obligation (see step 3b); never apply the equivalence check or its flags to a generalize verdict.
A maintainer flag fires only when the equivalence check itself turns up a problem, never merely because a term was checked:
- Equivalent present and agrees → record it, no flag.
- Equivalent missing →
D3b-parochialism.
- Equivalent present but conflicts →
naming-conflict or substantive-conflict (see the maintainer-flag vocabulary below).
The maintainer decides any flagged gap or conflict in governance review; this skill's obligation is to make sure they see it rather than having to watch non-US bodies by hand. Never resolve a missing or conflicting equivalent by inventing one — keep the grounded term and flag it.
Maintainer-flag vocabulary
A closed, controlled set of four tokens. Use exactly one of these (or none) per candidate — never a free-form note in the Maintainer flag column.
D3b-parochialism — no international/standards equivalent exists for the term (the term appears current only in one jurisdiction, or no crisp equivalent was found).
naming-conflict — an equivalent exists but recognized standards use a materially different label for the same concept (e.g. "reference monitor" vs. Common Criteria's "reference validation mechanism"). A label mismatch only — the underlying concept is the same.
substantive-conflict — same nominal concept, but the content or scope differs materially across standards (e.g. NIST AI RMF's seven trustworthiness characteristics vs. the EU HLEG's seven requirements — overlapping but not the same set). Distinct from naming-conflict, which is a label mismatch, not a scope/content mismatch.
missing-pin — the canonical source cannot be pinned to a traceable dated edition (no stable citation to anchor the term).
Informational surfacing that broadens the base without a defect (e.g. acknowledging EU/ISO framings that agree with or extend NIST) is not a maintainer flag — record it as informational context, not one of the four tokens.
Term verification (ADR-031 D4)
The lexicon holds stable canonical terms (PEP/PDP/PIP/PAP, reference monitor, least privilege) — treat these as a pinned snapshot. For a term or identifier that may have changed (a framework's evolving vocabulary), verify it against the source via WebSearch/WebFetch at use time rather than trusting the snapshot.
Output format
For each candidate term, report:
| Candidate | Verdict | Canonical term | Canonical source | International equivalent(s) | Maintainer flag |
|---|
- Verdict:
grounded · reground · generalize · novel-flag
- International equivalent(s): required for every
grounded/reground row — the recognized non-US/international standard(s) that cover the same concept, or "none found" when the equivalence check found nothing. For a generalize row, record "not applicable" — generalize carries no equivalence obligation (step 3b).
- Maintainer flag: one of the four controlled tokens (
D3b-parochialism, naming-conflict, substantive-conflict, missing-pin), or empty when the equivalence check found no problem. Never D3b-parochialism, naming-conflict, or substantive-conflict on a generalize row — that verdict is out of scope for the equivalence check entirely.
Close with the counterfactual list (proposed → chosen → source) for anything not already grounded.
Reference
references/lexicon.md — the canonical term set and the invented→canonical mappings, with sources.