Version 1.0Effective 2026-04-29

Capture.kit security one-pager

This page is what you forward to your InfoSec or procurement team. It is the single source of truth on what Capture.kit collects, what it doesn't, who the subprocessors are, and what the catalog's posture is on FedRAMP, CUI, and source-selection-sensitive material. A PDF version is at the bottom of the page for procurement files.

The headline

Capture.kit is not FedRAMP authorized.

The catalog ships productized AI assets — Prompt packs, Skills, Plugins, Agents — that the buyer runs inside their own Claude.ai or Claude Code environment. The catalog operates a marketing site, a Stripe checkout, a license-validation endpoint, and a magic-link customer dashboard. It does not operate a SaaS platform, does not centralize customer work product, and does not handle CUI, SSI, ITAR, or source-selection-sensitive material under any circumstance.

For workflows that require FedRAMP authorization, classified data handling, or CUI processing under DFARS 252.204-7012, this is the wrong product. A FedRAMP-aligned variant on AWS Bedrock GovCloud is on the roadmap as a separate product; this catalog is not it.

What runs where

The catalog is built around one architectural principle: the assets execute in your Claude environment, not ours.

A Prompt pack is a zip of text files. You paste them into a Claude.ai chat that runs on your Anthropic account. The catalog's servers are never in the loop while the prompt runs.

A Skill or Plugin is a .skill or .plugin bundle. You install it into your own Claude Projects, Claude Code, or Cowork environment. Anthropic's API serves the inference request from your account. The catalog's servers are not in the inference path.

The catalog's own infrastructure handles four narrow jobs: marketing pages, Stripe checkout, a one-call-per-day license-validation endpoint, and the magic-link customer dashboard. None of those touch your work product.

What we collect

Data	Why	Where stored	Retention
Email address	Order receipts, download links, magic-link login, lifecycle email	Postgres (Neon) + Beehiiv (broadcast list)	Until you unsubscribe and delete-account-request
Payment information	Charge for purchase	Stripe — we never see card numbers	Per Stripe's PCI-compliant retention
Order metadata	License generation, refund handling, support	Postgres (Neon)	Lifetime of the license + 7 years for tax
License validation pings	Verify active seat for prosumer and team-tier products. One ping at install + one per 24 hours for Plugins; one ping on install for Skills; no pings for Prompt packs.	Postgres (Neon) — license key, machine fingerprint hash, timestamp	Rolling 90 days for telemetry; license records kept for the license lifetime
Anonymous usage telemetry	Product analytics: page views, conversion funnel events. PostHog + Plausible.	PostHog + Plausible (privacy-respecting analytics)	Per provider defaults

DataEmail address
WhyOrder receipts, download links, magic-link login, lifecycle email
Where storedPostgres (Neon) + Beehiiv (broadcast list)
RetentionUntil you unsubscribe and delete-account-request
DataPayment information
WhyCharge for purchase
Where storedStripe — we never see card numbers
RetentionPer Stripe's PCI-compliant retention
DataOrder metadata
WhyLicense generation, refund handling, support
Where storedPostgres (Neon)
RetentionLifetime of the license + 7 years for tax
DataLicense validation pings
WhyVerify active seat for prosumer and team-tier products. One ping at install + one per 24 hours for Plugins; one ping on install for Skills; no pings for Prompt packs.
Where storedPostgres (Neon) — license key, machine fingerprint hash, timestamp
RetentionRolling 90 days for telemetry; license records kept for the license lifetime
DataAnonymous usage telemetry
WhyProduct analytics: page views, conversion funnel events. PostHog + Plausible.
Where storedPostHog + Plausible (privacy-respecting analytics)
RetentionPer provider defaults

What we do not collect

This is the part that matters most.

No RFP content. The Compliance Matrix Skill processes RFPs in your Claude environment. Capture.kit's servers never see the RFP, never see the matrix, never store any proposal-side artifact.
No proposal drafts. The Color Team Reviewer Agent runs Pink, Red, and Gold passes inside your Claude Code or Cowork session. Capture.kit's servers never see your draft, never see the findings, never store the rolled-up priority list.
No capture lists or pursuit pipelines. The SAM.gov Triage Pack runs in your Claude.ai chat. We never see your saved-search export, your capability profile, or your pursue/monitor/pass decisions.
No CUI, SSI, ITAR, source-selection-sensitive, or classified material. The AUP forbids uploading any of this to the catalog's tools. If you do, the violation is on you and the legal liability is on you.
No network telemetry from the asset runtime. Skills and Plugins do not phone home with proposal content, RFP excerpts, finding metadata, or any other artifact. The single network call is the license-validation ping, which carries the license key, a machine fingerprint hash, and the timestamp — nothing else.

Anthropic API posture

The catalog assets run on Anthropic's API in your account. The relevant terms:

Zero data retention is in effect for Anthropic's API by default for Enterprise customers, and is the explicit posture the catalog assumes. If your Claude account is on a tier where zero retention requires explicit configuration, configure it before running the assets on any sensitive work — even though the AUP already forbids sensitive content.
No training on your inputs. Anthropic does not train on Claude API inputs by default. The assets do not opt your inputs into training under any circumstance.
Your account, your costs. Inference cost lands on your Anthropic bill, not ours. Per-run cost ceilings are documented per product (≤$4 for Compliance Matrix; ≤$25 for Color Team Agent on a 200-page proposal).

Subprocessors

These are the third-party services Capture.kit uses to operate the marketing site, the checkout flow, and the license-validation endpoint. None of them are in the inference path or the asset-runtime path.

Vendor	Purpose	Data handled	Contract posture
Anthropic	Claude API runtime — runs inside the buyer's account, not ours	Inference inputs/outputs (the buyer controls)	Zero retention; no training on inputs
Stripe	Payments processing	Card data, billing address, tax ID	PCI Level 1; we do not store card data
Vercel	Marketing site hosting + serverless functions	Page-view request logs	SOC 2 Type II
Cloudflare	DNS, CDN, R2 file storage for download assets	Download requests, asset blobs	SOC 2 Type II
Neon	Postgres database — orders, licenses, magic-link tokens	Email, order metadata, license keys, fingerprint hashes	SOC 2 Type II; encrypted at rest
Resend	Transactional email — receipts, download links, magic-link login	Email address, receipt content, license key	SOC 2 Type II
Beehiiv	Broadcast and lifecycle email list	Email address, subscription preferences, content engagement	SOC 2 Type II
PostHog	Product event analytics	Anonymous event stream — page views, conversion events	SOC 2 Type II; no PII transmitted by design
Plausible	Privacy-respecting page analytics	Anonymous page visit counts	GDPR-compliant; no cookies, no tracking
Sentry	Error monitoring on serverless functions (Stripe webhook in particular)	Server-side error stacks	SOC 2 Type II

VendorAnthropic
PurposeClaude API runtime — runs inside the buyer's account, not ours
Data handledInference inputs/outputs (the buyer controls)
Contract postureZero retention; no training on inputs
VendorStripe
PurposePayments processing
Data handledCard data, billing address, tax ID
Contract posturePCI Level 1; we do not store card data
VendorVercel
PurposeMarketing site hosting + serverless functions
Data handledPage-view request logs
Contract postureSOC 2 Type II
VendorCloudflare
PurposeDNS, CDN, R2 file storage for download assets
Data handledDownload requests, asset blobs
Contract postureSOC 2 Type II
VendorNeon
PurposePostgres database — orders, licenses, magic-link tokens
Data handledEmail, order metadata, license keys, fingerprint hashes
Contract postureSOC 2 Type II; encrypted at rest
VendorResend
PurposeTransactional email — receipts, download links, magic-link login
Data handledEmail address, receipt content, license key
Contract postureSOC 2 Type II
VendorBeehiiv
PurposeBroadcast and lifecycle email list
Data handledEmail address, subscription preferences, content engagement
Contract postureSOC 2 Type II
VendorPostHog
PurposeProduct event analytics
Data handledAnonymous event stream — page views, conversion events
Contract postureSOC 2 Type II; no PII transmitted by design
VendorPlausible
PurposePrivacy-respecting page analytics
Data handledAnonymous page visit counts
Contract postureGDPR-compliant; no cookies, no tracking
VendorSentry
PurposeError monitoring on serverless functions (Stripe webhook in particular)
Data handledServer-side error stacks
Contract postureSOC 2 Type II

Acceptable Use Policy (summary)

The full AUP is at /aup. The summary that matters for procurement review:

Public data only. Customers may use the assets only on public solicitations, their own capability profiles, their own proposal drafts, and other content they have the right to process.
No CUI, SSI, ITAR, source-selection-sensitive, or classified content. Hard line. If you upload any of this, the AUP violation is yours and so is the legal liability.
Decision support only. The assets produce drafts, audits, and triage briefs. They do not make pursuit, bid, or submit decisions. The human reviewer remains the gate, and the [VERIFY] tag pattern is enforced in every output.
No re-distribution of the assets outside your licensed seats. Single-seat licenses are single-seat. Team licenses cover the seat count purchased.
No use of the assets to circumvent FAR or DFARS clauses in any solicitation or contract. The catalog is built to support compliance, not to evade it.

Compliance posture summary

| Framework | Status | Notes | |---|---|---| | FedRAMP | Not authorized | This catalog is explicitly outside FedRAMP. For FedRAMP workloads, wait for the GovCloud Bedrock variant. | | CMMC | Out of scope | The catalog does not handle CUI. CMMC scoping does not apply. | | DFARS 252.204-7012 | Out of scope | Same — no CUI processing. | | NIST 800-171 | Out of scope at the catalog layer | The buyer's Claude environment may carry 800-171 controls; the catalog itself does not handle controlled information. | | SOC 2 | In progress at the catalog layer | All upstream subprocessors carry SOC 2 Type II. The catalog's own SOC 2 posture is targeted for 2027. | | ISO 27001 | Out of scope at v1 | Roadmap consideration after SOC 2. | | GDPR / CCPA | Compliant at the marketing/checkout layer | Privacy policy at /legal/privacy. |

Vulnerability disclosure

Security issues should be reported to security@capture.kit (PGP key on file). Please do not open public issues for vulnerabilities. We respond within 48 hours and patch within 14 days for high-severity issues.

The catalog runs a continuous Sentry alert on the Stripe webhook handler — silent webhook failure is the catalog's worst-case operational failure mode, and we treat it that way.

Procurement contact

For security questionnaires, custom AUP language, NET-30 invoicing on team-tier subscriptions, or multi-seat purchase orders:

procurement@capture.kitResponse time: 48 hours

Do not send security questionnaires that assume FedRAMP authorization. The answer is no, and we'd rather you know that on day 0 than on day 30.

Download

Download as PDF (for your procurement file)240KB