--- name: compliance-claim-checker id: compliance-claim-checker version: "1.0.0" title: "Compliance Claim Checker" type: skill description: "Review documentation and PRs for federal compliance claims (FedRAMP, NIST, OWASP, Section 508) and flag uncited or overclaimed statements — without asserting compliance itself" status: experimental owners: - "@GSA-TTS/agentic-coding-team" primary_personas: - developers - security requires: anchors: [] output: format: markdown contract: required_sections: - "Summary" - "Findings" prohibited_content: - "Secrets" - "Real PII" - "Real CUI" - "Internal URLs" - "New Compliance Assertions" quality_gates: readability_max_grade: 10 citations_required: false triggers: - "compliance claim" - "FedRAMP" - "NIST" - "Section 508" - "authorized" - "compliant" - "overclaim" tags: - "security" - "compliance" - "documentation" - "citations" categories: - "security" - "compliance" - "documentation" risk_tier: low human_review_required: true allowed_tools: [] network_policy: deny write_policy: deny script_policy: deny portability: opencode: true cursor: true claude_projects: true chatgpt: true generic_llm: true scope: intended_use: - "Review docs, READMEs, and PR descriptions for compliance claims" - "Flag compliance statements that lack a citation to an authoritative source" - "Flag overclaiming (e.g. 'FedRAMP certified', 'NIST compliant') beyond what evidence supports" - "Point authors at the playbook's federal source registry for correct citations" exclusions: - "Does NOT determine or assert whether something is actually compliant" - "Not a substitute for an ATO, assessor, or authorizing official" - "Not legal advice; does not interpret regulation" - "Does not scan code for vulnerabilities (see secure-code-review)" source_inspiration: [] # No external source; grounded in the playbook's own registry. changelog: - version: "1.0.0" date: "2026-07-01" change_type: minor summary: "Initial version — flags uncited/overclaimed federal compliance statements, verifies citations against the playbook federal-ai-landscape registry, asserts no compliance itself." collection: security routing: task_types: - "review" - "analyze" input_artifacts: - "compliance-claim" - "documentation" - "pull-request-diff" output_artifacts: - "security-review" prefer_when: - "the request is to verify a stated FedRAMP/NIST/compliance claim" avoid_when: - "the request is a general code security review" aliases: - "compliance claim review" - "fedramp claim check" - "overclaim check" --- # Skill: Compliance Claim Checker Review documentation, READMEs, and pull-request text for **federal compliance claims** — statements asserting alignment with FedRAMP, NIST (SP 800-53, AI RMF, SSDF), OWASP, Section 508, or similar — and flag claims that are **uncited** or **overclaimed**. > **This skill checks claims; it does not certify compliance.** It never asserts > that anything *is* compliant, authorized, or certified. Authoritative policy > and control content live in the > [agentic-coding-playbook](https://github.com/GSA-TTS/agentic-coding-playbook); > this skill points authors there and defers all determinations to it. It is not > legal advice. ## When to Use - Reviewing a README, doc, or PR that claims compliance ("FedRAMP authorized", "meets NIST 800-53", "508 compliant", "OWASP Top 10 covered") - Before publishing federal-facing documentation that references standards - When an author cites a framework and you need to confirm the citation is real, current, and not overstated ## When NOT to Use - To *decide* whether a system is compliant — that is an assessor / authorizing official function, not this skill's - As evidence for an ATO package - To interpret a regulation's meaning (not legal advice) ## Prerequisites - Access to the text being reviewed (doc, README, or PR body) - Access to the playbook's federal source registry for citation verification: [`FEDERAL-AI-LANDSCAPE.md`](https://github.com/GSA-TTS/agentic-coding-playbook/blob/main/docs/FEDERAL-AI-LANDSCAPE.md) and its machine-readable source [`data/federal-ai-landscape.yaml`](https://github.com/GSA-TTS/agentic-coding-playbook/blob/main/data/federal-ai-landscape.yaml) ## Procedure ### 1. Extract compliance claims Scan the text for statements that assert alignment with, or authorization under, a standard or authority. Treat these claim-signal phrases as triggers: - Authorization/certification: "authorized", "certified", "accredited", "has an ATO", "FedRAMP authorized", "in the FedRAMP marketplace" - Conformance: "compliant with", "meets", "conforms to", "aligned with", "satisfies", "fully implements" - Framework names: FedRAMP, NIST SP 800-53, NIST AI RMF, NIST SSDF (800-218/218A), FISMA, Section 508, OWASP (LLM/Agentic/Top 10), OMB memoranda (M-YY-NN), ISO 42001 ### 2. Classify each claim For each extracted claim, classify it: | Class | Definition | |-------|------------| | **Cited & scoped** | Names a specific source AND scopes the claim (e.g. "implements the AC-6 least-privilege control per NIST SP 800-53") | | **Uncited** | Asserts alignment but names no source, control, or version | | **Overclaimed** | Asserts more than evidence supports — e.g. "certified"/"authorized"/"fully compliant" where only *alignment with practices* is shown, or a whole-framework claim from partial coverage | | **Stale/incorrect citation** | Cites a source that is superseded, rescinded, or misidentified | ### 3. Verify citations against the playbook registry For every cited framework/memo, confirm against the playbook's `federal-ai-landscape.yaml` registry that: - The source **exists** and is identified correctly (right document number/title) - Its **status** is current — flag citations to entries marked `rescinded`, `revoked`, or `superseded` (e.g. a doc citing a rescinded OMB memo as live guidance) - The claim doesn't misattribute a control or requirement to the wrong source If a cited source is **not in the registry**, do not assume it is wrong — flag it as "unverified against the registry" and note the registry may need the entry added (a tracker gap), rather than asserting the citation is invalid. ### 4. Distinguish "certified" from "aligned" Apply this bright line, which is the most common overclaim: - **"Certified" / "authorized" / "accredited" / "has an ATO"** are *status* claims granted by an external authority (a 3PAO, an AO, the FedRAMP PMO). A document MUST NOT assert these unless it cites the authorizing artifact. - **"Aligned with" / "implements the practices of" / "designed to meet"** are *effort* claims and are acceptable when scoped and cited. Recommend downgrading status claims to effort claims when no authorizing artifact is cited. ### 5. Report — never assert compliance Produce the output below. The report: - **Flags** problems and **suggests** citations or rewordings - **Never** states that the subject *is* compliant/authorized/certified - Points the author to the playbook for authoritative policy and to the registry for correct citations - When unsure whether a claim is overstated, flags it for human judgment rather than clearing it ## Output Contract ```markdown ## Summary <1-3 sentences: how many claims found, how many flagged, overall risk of overclaiming.> ## Findings For each flagged claim: - **Claim (quoted):** "" - **Location:** - **Class:** uncited | overclaimed | stale-citation | unverified-against-registry - **Why flagged:** - **Suggested fix:** | downgrade "certified" → "aligned with" | correct/refresh the citation | remove the claim> ## Notes - Registry gaps: - This review checks claims only; it does not determine compliance. See the playbook for authoritative policy. ``` ## Verification - Every flagged claim quotes the original text and names a class and a fix - The report makes **no** new compliance assertion of its own (check the output against the `New Compliance Assertions` prohibited-content rule) - Every "verify against the registry" step names the specific registry entry (or explicitly notes the source is absent from the registry) ## Examples ### Example 1 — overclaim (status vs effort) Input: `> This service is FedRAMP compliant and NIST 800-53 certified.` Finding: - **Class:** overclaimed - **Why:** "FedRAMP compliant" and "800-53 certified" are authorization/status claims with no cited authorizing artifact (ATO letter, FedRAMP marketplace listing). "Certified" is not a status NIST 800-53 confers. - **Suggested fix:** downgrade to "designed to align with NIST SP 800-53 Rev 5 controls" and, if an ATO exists, cite it; otherwise remove the FedRAMP status claim. ### Example 2 — uncited claim Input: `Our pipeline follows secure software development best practices.` Finding: - **Class:** uncited - **Suggested fix:** cite the source if a specific one is meant (e.g. "follows NIST SP 800-218 SSDF practices") or reword to avoid implying a standard. ### Example 3 — stale citation Input: `Per OMB M-24-10, our agency's AI governance…` Finding: - **Class:** stale-citation - **Why:** the registry marks M-24-10 as rescinded/superseded; the live OMB AI governance memo is M-25-21. - **Suggested fix:** cite M-25-21 as the current governance memo; reference M-24-10 only historically. ### Example 4 — well-formed claim (not flagged) Input: `Access control follows the least-privilege principle described in NIST SP 800-53 AC-6.` Finding: none — the claim is scoped (AC-6, least privilege) and cited (SP 800-53), and it describes an *effort/design* alignment, not a certification. ## References - Federal source registry (citation verification): [`FEDERAL-AI-LANDSCAPE.md`](https://github.com/GSA-TTS/agentic-coding-playbook/blob/main/docs/FEDERAL-AI-LANDSCAPE.md) - Authoritative policy (never restated here): [playbook `AGENTS.md`](https://github.com/GSA-TTS/agentic-coding-playbook/blob/main/AGENTS.md), [`SECURITY-CONTROLS.md`](https://github.com/GSA-TTS/agentic-coding-playbook/blob/main/docs/SECURITY-CONTROLS.md) - Governance model for security skills: [`docs/security-skill-governance.md`](../../../docs/security-skill-governance.md)