Skip to content

Latest commit

 

History

History
45 lines (30 loc) · 3.76 KB

File metadata and controls

45 lines (30 loc) · 3.76 KB

Intent vs Scope

Curbpack checks your repository against local rule packs and writes a review pack you can hand to a buyer or auditor—on your machine, without claiming certification.

Not conformity assessment. Not CE marking. Not a notified-body opinion.

Sixty-second clarity for buyers, auditors, and agents. Gate pass is evidence for human review — not conformity assessment, CE marking, or certification.

Authorities / CISO path: for-authorities. Voice: voice and terms. Abbreviations: glossary.

Column Content
Intent (why) Curbpack checks your repository against local rule packs and writes a review pack you can hand to a buyer or auditor—on your machine, without claiming certification. SMEs and suppliers need continuous, local, shareable evidence — without GRC SaaS or uploading IP to a cloud policy brain.
Curbpack scope (now) Pack JSON gates, check / heal, review-pack, CycloneDX / OpenVEX drafts, Git Notes attest, Action + SARIF, optional sock, explain-packets for tutors, export --lay-of-land / instrument.json, export --context-pack for assistants, thin MCP example, multi-IDE teach files (assistant-loop).
Not in scope (OSS) Conformity assessment, CE, OPA/Rego, LSP, syscall tracers, FIDO/EFOS, DNSSEC, cloud policy brain, LLM-as-judge, badge marketplace, gtm-oss on site, second pin, pack catalog growth before partner habit proof.
Pack catalog freeze Only house-policy, cra-baseline, medtech-iec62304 (ids). Enforced by scripts/redteam-pilot.sh allowlist; unlock only via freeze review + explicit PR (no CI env escape hatch).
v3.33 spec Internal R&D / EE north star only — not mirrored into OSS; not the adoption contract.
IP / chat boundary Raw source and secrets never leave the machine for “compliance chat.” Only sanitized GateFailure explain-packets may leave for an optional tutor the operator explicitly chooses.
Promotion bar ./scripts/redteam-pilot.sh green.

Local rule-pack checks (main flow)

repo → Curbpack (gates, packs, attest)
         ↓
   review-pack / exports (evidence for humans)
         ↓
   human judgment (attest when ready)

Curbpack decides pass/fail. Optional chat tutors (any local or operator-chosen assistant) may draft prose from a sanitized explain-packet — they never greenlight gates and never write attest capsules. After any proposed fix you must re-check (curbpack check). Optional sock sidecar: examples/mcp/README.md.

Agentic coding: evidence habit / not AI security product

Agents and humans share one loop: edit → curbpack check → on red heal/ask; on green optional --context-pack / --lay-of-land / --buyer-questions for humans. This is not an AI security product, SCA/CVE platform, or certification engine. Hooks keep agent PRs honest; tutors still require re-check before any “fixed” claim. See assistant-loop.

Compose with SCA / secret scanners

Curbpack prepares structural evidence for product repos. Pair it with SCA (e.g. Trivy/OSV) and secret scanners (e.g. Gitleaks) for depth. Explicitly: not a security program; watchlist∩SBOM is look-here only — not a CVE product.

Activate in under ten minutes: getting-started / 60-second paths.

See also: For authorities · Strategy boundary · Stable contracts · Promotion firewall · Write your own pack · Security model

Optional tutor product (integrators only): Coreward pointer.