TEMPLATE-CHECKLIST.md

SAF Technique Completion Checklist

Use this checklist with the technique template and the research protocol. A polished README is necessary but not sufficient: the research packet and its validation gates are part of a new technique contribution.

Contract and Classification

• technique-contract.yml defines one coherent adversary behavior.
• The adversary objective and crossed or abused trust boundary are explicit.
• In-scope and out-of-scope behavior is testable rather than aspirational.
• The closest existing SAF techniques were reviewed and distinguished.
• The candidate passes the admission rule: atomic behavior, agentic dependency or explicit profile, distinct boundary, and operational value.
• The proposed ID is unused, opaque, permanent, and its tactic assignment is justified.
• SAF Core and applicable domain profiles are assigned explicitly.
• Relationships use canonical types and have the required inverse records.
• The entry is present and consistent in research/framework-model.yml.

Claims and Evidence

• Material claims were inventoried before or during research, not reconstructed only after drafting.
• Every material claim has a stable claim_id, class, materiality, limitations, and status.
• Every supporting source has a stable source_id in research/source-manifest.yml.
• Every source was opened and reviewed beyond a snippet, abstract, or generated summary.
• Exact sections, pages, paragraphs, commits, lines, advisories, or stable headings are recorded.
• Inference, historical analogy, and direct MCP evidence are kept distinct.
• Conflicting evidence and unsuccessful source access are recorded.
• Two consecutive no-change passes support the documented saturation decision.
• The technique-level evidence status is justified for the complete behavior.

Technique Document

• Overview includes tactic, ID, research packet, documentation status, evidence status, severity rationale, observation status, and dates.
• Overview includes lifecycle status and applicable framework profiles.
• Scope defines the boundary, inclusions, exclusions, and distinguishing characteristics.
• Description states the objective, MCP-specific mechanism, boundary, and uncertainty.
• Attack Vectors identify delivery paths, affected components, and the boundary crossed.
• Technical Details include prerequisites, staged flow, and one safe end-to-end scenario.
• Evidence and Current State uses matching claim IDs and source IDs.
• Impact Assessment rates confidentiality, integrity, availability, and scope conditionally.
• Detection Methods name telemetry, fields, behavioral indicators, logic, false positives, limits, and tuning.
• Mitigation Strategies address the defining mechanism and link to valid SAF mitigation IDs.
• Related Techniques state both the relationship and the non-duplicate boundary.
• MITRE ATT&CK Mapping is labeled Direct or Analogous with a behavioral rationale.
• References contain every materially used source and its source ID.
• Version History records the contribution.

Detection and Validation

• detection-rule.yml contains the complete example analytic and a unique UUID.
• The README does not duplicate the complete detection rule.
• The analytic uses only telemetry and fields named in the technique.
• Positive, negative, boundary, and expected false-positive cases are tested when feasible.
• Detection maturity is recorded without treating syntax validity as evidence of effectiveness.
• Test commands and outcomes are recorded in quality-review.yml.
• If testing is infeasible, the contract and README contain a specific, reviewer-approved waiver.
• Detection limitations do not claim prevention or comprehensive coverage.

Rights, Safety, and Alignment

• publication-rights.yml covers every cited source and third-party artifact.
• Quotations, code, diagrams, datasets, trademarks, and adaptations have an explicit basis for use.
• Examples use inert domains, placeholder credentials, and non-destructive operations.
• Research-driven changes needed elsewhere are tracked in research/alignment-ledger.yml.
• No unresolved high-severity alignment or publication-rights issue remains.
• quality-review.yml records a passing result for every gate.
• python3 scripts/validate-technique-research.py SAF-TXXXX passes.
• python3 scripts/validate-framework-model.py passes the evidence, taxonomy, and operational release gates.
• python3 scripts/generate-technique-catalog.py --check confirms that the public catalog is current.

Directory Structure

techniques/
└── SAF-TXXXX/
    ├── README.md
    ├── detection-rule.yml
    ├── test-logs.json             # Required when representative testing is feasible
    └── test_detection_rule.py     # Or an equivalent validation command

research/
├── source-manifest.yml
├── framework-model.yml
├── alignment-ledger.yml
└── techniques/
    └── SAF-TXXXX/
        ├── technique-contract.yml
        ├── claim-inventory.yml
        ├── source-coverage.yml
        ├── publication-rights.yml
        └── quality-review.yml

Style Rules

• Use objective, mechanism-specific language.
• Prefer current primary sources, specifications, advisories, implementation artifacts, and peer-reviewed research.
• State uncertainty and preconditions near the affected claim.
• Use canonical identifiers such as SAF-T1001, SAF-M-1, and ATK-TA0001.
• Use Not observed in production when no directly reviewed source establishes real-world MCP or agentic-system exploitation.
• Do not call generic behavior an indicator of compromise.
• Do not treat a valid Sigma document as proof that its analytic is effective.