Skip to content

SKILL.md

Latest commit

 

History

History
360 lines (256 loc) · 22.3 KB

SKILL.md

File metadata and controls

360 lines (256 loc) · 22.3 KB
Raw
name build-threat-catalog
description Create a CCC threat catalog (threats.yaml) for a cloud service — import applicable core threats, define service-specific threats mapped to the service's capabilities, and map each to external frameworks (MITRE ATT&CK, MITRE D3FEND, CISA KEV, CWE, OWASP) via standalone Gemara MappingDocuments in the service's mappings/ directory, validating against schemas/threats-schema.json and schemas/mappingdocument-schema.json. Requires the service folder to already contain metadata.yaml and capabilities.yaml. Use when the user asks to "identify threats for `service`", "create a threat catalog for `service`", or "create threats.yaml" for a service.

Threat Catalog Skill

Purpose

Identify and create security threats for a cloud service, supporting the onboarding process for new cloud service threats in the CCC repository. This is the bridge between capabilities and controls: capabilities define what the service can do, threats define what can go wrong, and controls mitigate the threats.

Final Outcome

A threats.yaml file created in the service folder that imports applicable core threats from catalogs/core/core/threats.yaml, defines service-specific threats mapped to the service's capabilities, and grounds each threat in current adversary and exploitation evidence, validated against schemas/threats-schema.json — plus one standalone Gemara MappingDocument per external framework in the service's mappings/ directory (mappings/threats-<framework>.yaml, validated against schemas/mappingdocument-schema.json) connecting the threats to adversary techniques (MITRE ATT&CK), defensive countermeasures (MITRE D3FEND), known exploited vulnerabilities (CISA KEV), and weakness taxonomies (CWE, OWASP). Threats never carry inline external-mappings; that field is not part of the schema.

When to Use

When the user asks to identify or create threats for a cloud service. For example, "Identify threats for service", "Create a threat catalog for service", or "Create threats.yaml for the service in path".

Prerequisites

The target service folder must already contain:

  • metadata.yaml (provides the service abbreviation, CSP service links, and mapping-references)

  • capabilities.yaml (provides the capability surface threats are mapped against)

If either file is missing, stop and instruct the user to run the Capability Catalog skill first, since every threat must map to at least one capability.

Reference Sources

These four sources feed threat work along two distinct axes. Discovery sources inform what threats exist and whether they are realistic; they are consulted during Step 3/4 reasoning and need not appear in the YAML. Mapping sources become standalone MappingDocuments in the service's mappings/ directory (one document per framework, Step 6) and are gated by metadata.mapping-references (see the gate rule below).

Source reference-id Role Entry ID format Canonical source
MITRE ATT&CK (Enterprise — cloud platforms: IaaS, SaaS, Identity Provider, Office Suite) MITRE-ATT&CK Discovery + Mapping Txxxx, Txxxx.xxx attack.mitre.org
MITRE D3FEND (v1.x knowledge graph of countermeasures) D3FEND Mapping (threat → countermeasure bridge to controls) D3-XXX (e.g. D3-NTA) d3fend.mitre.org
CISA Known Exploited Vulnerabilities Catalog CISA-KEV Discovery + Mapping (sparingly, as exploitation evidence) CVE-YYYY-NNNNN cisa.gov/known-exploited-vulnerabilities-catalog; JSON feed at https://www.cisa.gov/sites/default/files/feeds/known_exploited_vulnerabilities.json (GitHub mirror: cisagov/kev-data)
CSP Security Advisories (AWS / Azure / GCP) remarks-level by default, or CSP-Advisory if declared Discovery + Evidence provider-specific (AWS-YYYY-NNN, MSRC ADV/CVE, GCP-YYYY-NNN) AWS: aws.amazon.com/security/security-bulletins · Azure: msrc.microsoft.com + service bulletins on learn.microsoft.com · GCP: cloud.google.com/support/bulletins
CWE / OWASP Top 10 (unchanged) CWE, OWASP-Top-10 Mapping CWE-nnn, Axx:YYYY cwe.mitre.org · owasp.org

Source-role notes — read before mapping:

  • MITRE ATT&CK is the primary discovery lens for adversary behaviour. Prefer techniques drawn from the Enterprise cloud platforms over generic endpoint techniques, and prefer a specific sub-technique (Txxxx.xxx) over its parent when one applies.

  • MITRE D3FEND describes countermeasures, not threats, so it is the forward-link to the Control Catalog rather than a description of the risk itself. Derive D3FEND techniques from a threat's ATT&CK mapping (D3FEND maps countermeasures to the ATT&CK techniques they counter via the Digital Artifact Ontology) and record them as "the defensive technique classes that address this threat." Treat D3FEND mappings as optional; omit when no ATT&CK anchor exists.

  • CISA KEV is product- and CVE-specific, whereas CCC threats are provider-neutral and capability-based. Use KEV mainly to confirm realizability (Step 4) — if a weakness class in this service category appears in KEV, the threat is demonstrably live. Emit a CISA-KEV external mapping only when a threat corresponds directly to a class of weakness with KEV entries, cite one or two representative CVE IDs, and make the remarks state that the CVE is illustrative evidence (with the affected product and dateAdded). Never invent a 1:1 threat→CVE mapping to fill the block.

  • CSP Security Advisories are not a stable framework with durable mapping IDs. By default they are a discovery and grounding source: use them in Step 4 to find documented, real-world failure modes for each capability across AWS, Azure, and GCP, and cite the relevant advisory in a threat's remarks. Only emit a structured CSP-Advisory external mapping if the team has explicitly declared CSP-Advisory in metadata.mapping-references.

Mapping-reference gate. A framework may get a MappingDocument only if its reference-id is declared in metadata.mapping-references. If MITRE-ATT&CK, D3FEND, or CISA-KEV is not yet declared:

  1. Still use it for discovery and realizability reasoning, and for remarks.

  2. Do not create a mapping document for it.

  3. Surface a recommendation in the Step 4 confirmation block to add the missing reference-id(s) to metadata.mapping-references (this is the metadata/Capability Catalog skill's responsibility, not this skill's — do not edit metadata.yaml here).

Step 1: Locate Service and Validate Prerequisites

  1. Request the target service folder path (e.g., catalogs/storage/object/).

    • If no path is given, ask for the service name and resolve the folder under catalogs/.
    • If the folder cannot be resolved, list candidate folders and ask for clarification.

  2. Verify the prerequisite files exist in the folder.

  3. Read metadata.yaml and extract:

    • The service abbreviation from metadata.id (e.g., ObjStor from CCC.ObjStor).
    • The example-csp-services entries (AWS, Azure, GCP names and documentation links).
    • The mapping-references list — only these frameworks may receive MappingDocuments later.

  4. Check whether a threats.yaml already exists in the folder and note it.

  5. Compare the declared mapping-references against the Reference Sources table. Note which of MITRE-ATT&CK, D3FEND, and CISA-KEV are declared (usable as mappings) versus undeclared (discovery-only, per the mapping-reference gate).

Output Format

First output line must be: Step 1: Locate Service and Validate Prerequisites

Return the validation result in this format:

Target path: <catalogs/.../...> Service ID: CCC.<ABBREVIATION> Prerequisites: metadata.yaml: found|missing | capabilities.yaml: found|missing Existing threats.yaml: yes|no Mapping references (declared): MITRE-ATT&CK, D3FEND, CISA-KEV, CWE, ... Mapping sources usable: list declared | Discovery-only (undeclared): list undeclared Confidence: High|Medium|Low

Do not proceed to Step 2 if any prerequisite is missing.

Step 2: Capability Surface Review

  1. Read capabilities.yaml from the service folder and build the full capability inventory:

    • imported-capabilities entries (core capabilities, e.g., CCC.Core.CP11) with their remarks.
    • Service-specific capabilities entries (e.g., CCC.<ABBREVIATION>.CP01) with title and group.

  2. Each capability represents an attack surface. Every service-specific threat must map to at least one capability it puts at risk. Track this mapping explicitly.

  3. Review the official AWS, Azure, and GCP documentation links from metadata.yaml to understand how each capability is exposed and where misconfiguration or abuse is feasible.

Output Format

First output line must be: Step 2: Capability Surface Review

Return the capability inventory in a markdown table:

Capability ID Title Source Group
CCC.Core.CP11 title imported group
CCC.<ABBR>.CP01 title service-specific group

Total capabilities forming the attack surface: n Confidence: High|Medium|Low

Step 3: Core Threat Reuse

  1. Read catalogs/core/core/threats.yaml and review all core threats (CCC.Core.TH*).

  2. Select core threats for import when the threat applies to one or more of the service's capabilities (e.g., import CCC.Core.TH02 "Data is Intercepted in Transit" only if the service transmits data over the network).

  3. Do not plan service-specific threats that duplicate the intent of an imported core threat. If a core threat such as CCC.Core.TH01 (Access Control is Misconfigured) already covers the risk generically, keep it in imported-threats only and reserve service-specific threats for risks unique to this service's behavior.

  4. Compare against a peer catalog in the same category (e.g., catalogs/storage/object/threats.yaml for a storage service) as a sanity check on which core threats are conventionally imported.

Output Format

First output line must be: Step 3: Core Threat Reuse

Return the selected imports in a markdown table:

Core Threat ID Title Applies To Capability
CCC.Core.TH02 title CCC.<ABBR>.CP03

Confidence: High|Medium|Low

Step 4: Service-Specific Threat Identification

4a. Source-driven discovery

Before drafting threats, work each capability from the Step 2 inventory against the discovery sources in the Reference Sources table. This grounds the catalog in real adversary behaviour and demonstrated exploitation rather than speculation:

  1. MITRE ATT&CK — for each capability, identify the adversary techniques feasible on that surface, drawing from the Enterprise cloud platforms (IaaS, SaaS, Identity Provider, Office Suite). These technique IDs seed both the threat statement and its MITRE-ATT&CK mapping.

  2. CISA KEV — check whether weaknesses in this service category (or its common implementation patterns) appear in the KEV catalog. A KEV hit is strong evidence that the threat is realizable in the wild and should raise the threat's priority; record representative CVE IDs and their dateAdded for use in remarks or, where the gate allows, a sparing CISA-KEV mapping.

  3. CSP Security Advisories — review the AWS, Azure, and GCP security bulletins for the services behind each capability (see Reference Sources for URLs) to surface documented, provider-side failure modes. Use these to make threats concrete and provider-neutral (a real issue on one CSP usually points to an analogous risk on the others). Cite the advisory in the threat's remarks.

  4. MITRE D3FEND — for each candidate threat's ATT&CK techniques, note the countermeasure technique classes that counter them. This is forward-looking toward the Control Catalog and seeds the optional D3FEND mapping.

4b. Threat drafting

  1. Identify risks that arise from this service's distinct behavior and are not already covered by an imported core threat.

  2. For each capability in the Step 2 inventory, consider the failure modes feasible on that surface across AWS, Azure, and GCP, informed by the discovery sources above. Prefer granular, service-specific threats over broad umbrella statements when the behavior is meaningfully distinct (e.g., separate "Dead-Letter Queue Exposes Sensitive Payloads" from "Messages Replayed by Unauthorized Consumers").

  3. Each proposed threat must:

    • Map to at least one capability from the Step 2 inventory.
    • Be realizable on all three CSPs (provider-neutral), even if the mechanism differs. Evidence from a single CSP advisory or a CVE in KEV satisfies the realizability check; generalize the underlying weakness to a provider-neutral statement.
    • Use a group id defined in catalogs/core/core/groups.yaml (e.g., Encryption, Access, Observability, Data, Resource).

  4. Number threats sequentially: CCC.<ABBREVIATION>.TH01, CCC.<ABBREVIATION>.TH02, ...

    • If updating an existing threats.yaml, continue numbering after the highest existing id and do not renumber existing threats.

  5. Follow style-guides/catalogs/threat-style-guide.yaml for titles and descriptions:

    • Titles: ≤12 words, title case, framed as a negative event or security failure (e.g., "Data is Exposed to Unauthorized Consumers").
    • Descriptions: multi-line | text with a three-part structure — Circumstances (conditions/mechanism), Effect (what happens to the system), Impact (effect on confidentiality, integrity, or availability).
    • Use present tense and passive voice when describing manifestation ("may be misconfigured", "could be exploited"). Avoid intent-based words (accidental, malicious, deliberately) and speculative hedging (might possibly). Focus on technical mechanisms, not attacker motivation.
    • Use the precise vocabulary from the style guide: user, component, child resource, external system.

  6. Identify external framework mappings for each threat, observing the mapping-reference gate (a framework may get a mapping document only if its reference-id is declared in metadata.mapping-references). These feed the MappingDocuments written in Step 6:

    • MITRE-ATT&CK — technique IDs (Txxxx / Txxxx.xxx) from Step 4a. Prefer cloud-platform techniques and specific sub-techniques.
    • D3FEND — defensive technique IDs (D3-XXX) that counter the threat's ATT&CK techniques. Optional; the bridge to the Control Catalog. Omit when there is no ATT&CK anchor.
    • CISA-KEV — CVE-YYYY-NNNNN IDs, used only as illustrative exploitation evidence for a matching weakness class. remarks must state the affected product and dateAdded and note the CVE is illustrative. Use sparingly; never force a mapping.
    • CWE / OWASP-Top-10 — as before.
    • CSP advisories — cite in threat remarks by default; create a CSP-Advisory mapping document only if CSP-Advisory is declared in metadata.mapping-references.
    • Skip a framework entirely when no confident mapping exists rather than guessing.

Output Format

First output line must be: Step 4: Service-Specific Threat Identification

Return the proposed threats in a markdown table:

Threat ID Group Title Maps To Capabilities ATT&CK D3FEND KEV / Other Evidence (source)
CCC.<ABBR>.TH01 group title CCC.<ABBR>.CP01 T1020 D3-OTF CVE-2024-XXXXX, CWE-200 GCP bulletin GCP-2024-NNN

(Leave a cell blank where no confident mapping or evidence exists. Frameworks shown in ATT&CK/D3FEND/KEV columns that are not declared in metadata.mapping-references are discovery-only and must not get a mapping document in Step 6 — they remain in remarks instead.)

At the end of Step 4, return a single confirmation block in this format:

Service ID: CCC.<ABBREVIATION> Imported core threats: n Service-specific threats: n Capabilities referenced: n of total Mapping frameworks used: e.g. MITRE-ATT&CK, D3FEND, CWE Discovery-only frameworks (not declared in metadata): list, or none Recommended metadata.mapping-references additions: list, or none Target files: <catalogs/.../.../threats.yaml> plus mappings/threats-<framework>.yaml per framework

Reply with one of the following:

CONFIRM EDIT

Do not proceed to Step 5 until the user replies CONFIRM. If the user replies EDIT, apply the edits and return the updated Step 4 confirmation block, then wait for CONFIRM.

Step 5: Create threats.yaml

  1. Use schemas/threats-schema.json as the source of truth for required and allowed fields.

  2. Build imported-threats from the confirmed Step 3 selection using the schema shape:

    imports:
  - reference-id: CCC
    entries:
      - reference-id: CCC.Core.TH01
        remarks: <core threat title>

  3. Build threats from the confirmed Step 4 list. Each threat must include id, title, description, and group:

    threats:
  - id: CCC.<ABBREVIATION>.TH01
    group: <Group>
    title: <Threat Title>
    description: |
      <Circumstances — the conditions or mechanism>. <Effect — what
      happens to the system>. <Impact — effect on confidentiality,
      integrity, or availability>.
    capabilities:
      - reference-id: CCC
        entries:
          - reference-id: CCC.<ABBREVIATION>.CP01
            remarks: <capability title>
          - reference-id: CCC.Core.CP11
            remarks: <core capability title>

    Threats never carry an inline external-mappings field — external framework references belong in the Step 6 MappingDocuments.

  4. Capability mapping rules:

    • Every threat must reference at least one capability from the Step 2 inventory.
    • reference-id capability IDs must match the pattern CCC[.<service>].CP<n> and exist in the service capabilities.yaml or core capabilities.
    • Include remarks with the capability title for readability.

  5. Validate the final object against schemas/threats-schema.json before writing the file. Verify:

    • Every group id exists in catalogs/core/core/groups.yaml.
    • Every capability reference-id exists in the service capabilities.yaml or core capabilities.

  6. Write the file to <target-path>/threats.yaml.

  7. If a threats.yaml already exists, show a diff-style summary and ask for confirmation before overwrite.

Output Format

First output line must be: Step 5: Create threats.yaml

Return the threats creation result in this format:

Threats File: <catalogs/.../.../threats.yaml> Threats Status: created|updated|pending-confirmation Imported Threats: n | Service-Specific Threats: n Capabilities Referenced: n of total Validation: passed|failed Confidence: High|Medium|Low

Step 6: Create Mapping Documents

External framework references are standalone Gemara MappingDocuments — one file per target framework — in the service's mappings/ directory. Use schemas/mappingdocument-schema.json as the source of truth for required and allowed fields, and an existing document (e.g. catalogs/storage/object/mappings/threats-mitre-attack.yaml) as a reference.

  1. For each framework confirmed in Step 4 (and passing the mapping-reference gate), create <target-path>/mappings/threats-<framework-slug>.yaml with a lowercase framework slug (e.g. threats-mitre-attack.yaml, threats-d3fend.yaml, threats-cwe.yaml):

    title: <Service Title> Threats to <Framework>
metadata:
  id: CCC.<ABBREVIATION>.TH.threats-to-<framework-slug>
  type: MappingDocument
  gemara-version: 1.2.0
  version: dev
  description: Maps <service> threats to <framework> entries they relate to.
  author:
    id: FINOS-CCC
    name: FINOS Common Cloud Controls
    type: Human
  mapping-references:
    - id: CCC.<ABBREVIATION>.TH
      title: <Service Title> Threats
      version: dev
    - id: <framework reference-id, e.g. MITRE-ATT&CK>
      title: <framework title>
      version: <framework version, or dev_to-be-determined>
      url: <framework canonical URL>
source-reference:
  reference-id: CCC.<ABBREVIATION>.TH
  entry-type: Threat
target-reference:
  reference-id: <framework reference-id>
  entry-type: <Vector for attack techniques; per WG convention otherwise>
mappings:
  - id: CCC.<ABBREVIATION>.TH01-<framework-slug>
    source: CCC.<ABBREVIATION>.TH01
    relationship: relates-to
    targets:
      - entry-id: T1020
        remarks: <entry title, or evidence notes where the framework rules require them>

  2. Mapping rules (carried over from Step 4):

    • One mappings[] entry per threat that maps to the framework; threats with no confident mapping simply do not appear.
    • relationship defaults to relates-to unless the working group has ratified a stronger relationship for the pair.
    • MITRE-ATT&CK: technique IDs from the cloud platforms; sub-techniques preferred over parents.
    • D3FEND: countermeasure IDs (D3-XXX) derived from the threat's ATT&CK techniques; remarks should note which technique each counters. Skip the document if no threat has an ATT&CK anchor.
    • CISA-KEV: CVE IDs as illustrative exploitation evidence only; remarks must include the affected product and dateAdded and flag the CVE as illustrative. Do not fabricate a CVE. The threat statement itself stays CVE-free — KEV lives only in the mapping document.
    • CSP advisories: record in the threat remarks by default; create a CSP-Advisory document only when that reference-id is declared in metadata.

  3. Validate each document against schemas/mappingdocument-schema.json before writing. Verify every source threat id exists in the Step 5 threats.yaml and both source-reference and target-reference ids appear in the document's own metadata.mapping-references.

  4. If a mapping document already exists for a framework, update it in place (continue id numbering from the threat ids; never renumber) and show a diff-style summary before overwrite.

Output Format

First output line must be: Step 6: Create Mapping Documents

Return the mapping documents result in this format:

Mapping documents: catalogs/.../.../mappings/threats-<slug>.yaml, ... Documents Status: created|updated|pending-confirmation Mappings emitted: MITRE-ATT&CK: n | D3FEND: n | CISA-KEV: n | CWE: n | ... Frameworks skipped (not in mapping-references): list, or none Validation: passed|failed Confidence: High|Medium|Low