thanhnvandClaude Fable 5
da66a36f97
Close the three gaps the scoring report itself flagged for H6 plus V15, each as a real MVP + fail-able adversarial test (same pattern that lifted H5): - D1 alert-dispatch.sh: alerts POST to a real HTTP webhook (severity routing, dedup window, retry) + dead-letter queue with redelivery; fail-loud in strict. Wired into agent-metrics.sh so a failing step pages live end-to-end. - D2 provider-usage-fetch.sh + telemetry-reconcile.sh: pull usage from a provider usage HTTP API (all-or-nothing schema gate, fail-loud) + reconcile local vs provider ground truth — token under-reporting/hidden runs => TELEMETRY_DISCREPANCY. - D3 dashboard-serve.sh + dashboard-server.py: serve the dashboard over HTTP with a stale-aware /healthz probe (fresh=200 ok, telemetry silent-death=503 stale). - D4 circuit-breaker-check.sh: sliding-window failure-rate breaker (V15) — interleaved successes no longer evade the consecutive-failure breaker (CIRCUIT_OPEN_WINDOW). New suite phase-h6-agentops-tests.sh: 20/20, all live against local HTTP endpoints (webhook sink, mock provider API, dashboard server) — deterministic, no model needed. Also fix sign-policy-bundle.sh key-sync invariant: the local-fallback branch only exported policy-public.pem when generating a NEW key, so a Vault-DOWN run after a Vault-signed run verified a local-key signature against the Vault pubkey (RSA padding error, run-casan4 died mid-suite). Now always re-exports the pubkey before signing — same fix class as tool-audit-lib.sh / governance-check.sh. Full battery re-run sequentially: 175/175 PASS, 0 FAIL across 8 suites (KMS SKIP this run — Vault down; validated live 2026-07-04). Docs synced: scoring-run-report (H6 79→80, no harness below 80, 155→175), CASAN_HARDENING_STATUS (Phase 5 D1–D4), Plan-07, submission README, and run-hardening.sh (H6+ scenes HO1–HO4). Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
105 lines
10 KiB
Markdown
105 lines
10 KiB
Markdown
# CASAN Production Hardening — Implementation Status
|
||
|
||
> Live record of what has been **implemented and tested** from Plan-07
|
||
> (Production Hardening) and Plan-09 (Evidence Pack), versus what is still
|
||
> **planned**. Honest by design: a control is only listed as done when it has an
|
||
> executable test that would fail if the control were removed.
|
||
>
|
||
> Labels: **[implemented+tested]** · **[scaffold+tested]** (works, but a stronger
|
||
> production form is documented as TODO) · **[planned]** (not built).
|
||
|
||
## 1. What is implemented now
|
||
|
||
### Phase 1 — Plan-07 Track A (low-risk hardening) — [implemented+tested]
|
||
|
||
| ID | Control | Where | Test |
|
||
|---|---|---|---|
|
||
| A1 | H4 semantic **strict fail-closed** (`CASAN_SECURITY_STRICT=1`): model unavailable ⇒ BLOCK, never silent SKIP; non-strict logs `SEMANTIC_SKIPPED` loudly | `security-check.sh` | phase1 A1 |
|
||
| A2 | H4 **unicode/encoding normalization**: NFKC + zero-width strip + homoglyph fold + base64/hex decode & rescan | `security-check.sh`, `unicode-normalize.py`, `decode-suspicious.py` | phase1 A2 |
|
||
| A3 | H4 **tool-output injection scan** before output re-enters model context (mode off/warn/block; strict⇒block) | `tool-output-scan.sh`, `casan-harness.sh` | phase1 A3 |
|
||
| A4 | H5 **telemetry integrity**: provider-usage + cost metrics bound to a signed manifest head (tamper ⇒ mismatch; re-forge ⇒ signature invalid) | `telemetry-integrity.sh`, `sign-audit-head.sh` | phase1 A4 |
|
||
| A5 | H6 **cost controls**: absolute per-call cap + cumulative budget + cold-start protection, keeping median×mult spike | `cost-spike-detect.sh` | phase1 A5 |
|
||
| A6 | **Benign corpus (EN/VI/JA, 95 samples) + FP budget gate**: FP ≤ 3%, adversarial block ≥ 95%, CRITICAL = 100% | `benign-fp-report.sh`, `benign-corpus/`, `redteam-vectors.jsonl` | phase1 A6 |
|
||
|
||
### Phase 2 — Plan-07 Track C-MVP (production minimum bar) — mixed
|
||
|
||
| ID | Control | Status | Where | Test |
|
||
|---|---|---|---|---|
|
||
| C1 | **Tool authorization / action gating** (ALLOW/WARN/REQUIRE_APPROVAL/BLOCK): sensitive-file writes, destructive/remote-exec commands, dep-install & network egress | [implemented+tested] | `action-gate.sh` | phase2 C1 |
|
||
| C2 | **Supply-chain gate**: manifest diff, denylist + typosquat + dangerous lifecycle-script block, new-dep approval, dep-diff report | [implemented+tested] | `supply-chain-gate.sh`, `supply-chain-scan.py` | phase2 C2 |
|
||
| C3 | **Data-exfil guard**: secret-to-cloud & artifact-leaks-env ⇒ BLOCK; PII-in-audit ⇒ mask | [implemented+tested] | `data-exfil-guard.sh` | phase2 C3 |
|
||
| C6 | **Runtime sandbox**: static policy block (read ~/.ssh, net egress, fork bomb, write-outside-workspace, huge-file) + ulimit/timeout backstops | [scaffold+tested] | `sandbox-run.sh` | phase2 C6 |
|
||
|
||
### Phase 3 — Plan-09 Evidence Pack (MVP) — [implemented+tested]
|
||
|
||
| Control | Where | Test |
|
||
|---|---|---|
|
||
| `casan pack` / `casan verify-pack` (mapped to `evidence-pack.sh`): standard 12-file pack, hash manifest, signed head, tamper-evident verify, certified-run gate (no false certification, no silent skip) | `evidence-pack.sh`, `evidence-pack-build.py`, `evidence-pack-verify.py` | phase3-evidence-pack |
|
||
|
||
### Phase 4 — H5 governance hardening (raises the lowest harness) — mixed
|
||
|
||
| ID | Control | Status | Where | Test |
|
||
|---|---|---|---|---|
|
||
| C4 | **Approval-identity**: high-risk approval trusted only when a REGISTERED reviewer cryptographically SIGNS the request and their role is authorized — env-var approver no longer enough (SoD still enforced) | [implemented+tested] | `approval-sign.sh`, `approval-verify.sh`, `reviewers.registry`, `governance-check.sh` (`CASAN_APPROVAL_STRICT=1`) | phase-h5-approval (8) |
|
||
| B3 | **KMS key management**: sign audit/telemetry head via Vault Transit (key `exportable:false` → never leaves KMS) + key **rotation**; validated live | [implemented+tested] (live when Vault present; skip-aware otherwise) | `vault-kms.sh` (`rotate`, `assert-nonexportable`), `sign-audit-head.sh` | phase-h5-infra (KMS) |
|
||
| C5 | **External WORM audit**: ship audit head to a hash-linked append-only ledger; detect local rollback (`AUDIT_GAP_DETECTED`) and ledger tamper (`AUDIT_LEDGER_TAMPERED`) | [implemented+tested] (local ledger MVP) | `worm-ledger.py`, `audit-ship.sh`, `verify-audit-gap.sh` | phase-h5-infra (WORM) |
|
||
|
||
### Phase 5 — H6 AgentOps hardening (raises the lowest harness) — mixed
|
||
|
||
| ID | Control | Status | Where | Test |
|
||
|---|---|---|---|---|
|
||
| D1 | **Live alert dispatch**: alerts POST to a real HTTP webhook (severity routing, dedup window, retry); undelivered alerts fail-loud (`ALERT_DELIVERY_FAILED`, strict) and queue to a **dead-letter** file with redelivery (`--flush-deadletter`); wired into `agent-metrics.sh` (failing step ⇒ live page) | [implemented+tested] (live local webhook; production points at Slack/PagerDuty + on-call) | `alert-dispatch.sh`, `agent-metrics.sh` | phase-h6-agentops (①) |
|
||
| D2 | **Provider-telemetry API + reconciliation**: pull usage from a provider usage HTTP API (all-or-nothing schema gate, `PROVIDER_API_UNREACHABLE`/`PROVIDER_USAGE_INVALID` fail-loud) + reconcile local metrics vs provider ground truth — under-reporting/hidden runs ⇒ `TELEMETRY_DISCREPANCY` | [implemented+tested] (live local API endpoint; production points at OpenAI/Anthropic usage APIs) | `provider-usage-fetch.sh`, `telemetry-reconcile.sh` | phase-h6-agentops (②) |
|
||
| D3 | **Hosted dashboard**: dashboard served over HTTP with a stale-aware `/healthz` probe (fresh ⇒ 200 ok; telemetry silent-death ⇒ 503 stale — page-able by any uptime monitor) | [implemented+tested] (local HTTP daemon; production host = nginx/container, same routes) | `dashboard-serve.sh`, `dashboard-server.py` | phase-h6-agentops (③) |
|
||
| D4 | **Sliding-window circuit breaker (V15)**: failure **rate** over the last N calls trips `CIRCUIT_OPEN_WINDOW` — interleaving successes no longer evades the consecutive-failure breaker | [implemented+tested] | `circuit-breaker-check.sh` | phase-h6-agentops (④) |
|
||
|
||
## 2. Test inventory (all suites)
|
||
|
||
| Suite | Checks | Purpose |
|
||
|---|---|---|
|
||
| `run-casan4-harness-tests.sh` | 35 | Baseline happy-path + Level-5 evidence (unchanged) |
|
||
| `adversarial-harness-tests.sh` | 44 | Original adversarial battery (unchanged) |
|
||
| `phase1-track-a-tests.sh` | 25 | Track A hardening |
|
||
| `phase2-track-c-tests.sh` | 29 | Track C-MVP |
|
||
| `phase3-evidence-pack-tests.sh` | 7 | Evidence Pack MVP |
|
||
| `phase-h5-approval-tests.sh` | 8 | Approval-identity (C4) |
|
||
| `phase-h5-infra-tests.sh` | 7 | KMS (B3, live/skip-aware) + WORM (C5) |
|
||
| `phase-h6-agentops-tests.sh` | 20 | **New** — live alerting (D1) + provider-API/reconcile (D2) + hosted dashboard (D3) + window breaker (D4); all against live local HTTP endpoints |
|
||
| **Total** | **175** | Baseline 79 preserved; +96 new hardening checks. Last full run 2026-07-05 @ head of `feat/plan07-track-a-hardening`, 0 fail (KMS SKIP this run — validated live 2026-07-04 via Vault dev). |
|
||
|
||
Run order note: `run-casan4-harness-tests.sh` does `rm -rf .specify/logs`, so run it
|
||
**first** and never concurrently with the other suites.
|
||
|
||
## 3. What is NOT done (still planned — do not claim as production-ready)
|
||
|
||
| Area | Status | Plan ref |
|
||
|---|---|---|
|
||
| H4 multilingual detection (VI/JA injection block-patterns) | [planned] | Plan-07 B1 (V2) |
|
||
| Classifier-inject / split-injection resistance | [planned] | Plan-07 B2 (V5,V6) |
|
||
| Model-digest pinning | [planned] — sliding-window circuit breaker (V15) is now done (Phase 5 D4) | Plan-07 B4 (V16) |
|
||
| Live alerting to a managed channel (Slack/PagerDuty + on-call rota) | [partial] — webhook dispatch + dedup + dead-letter done; managed channel & escalation are config away, incident workflow is C7 | Plan-07 C7 / Phase 5 D1 |
|
||
| Hosted telemetry dashboard | [partial] — HTTP-served dashboard + stale-aware `/healthz` done locally; deployed host (nginx/container, auth) planned | Phase 5 D3 |
|
||
| Provider billing-API telemetry | [partial] — API fetch + schema gate + local-vs-provider reconciliation done against a live local endpoint; real OpenAI/Anthropic usage-API calls (needs keys) planned | Phase 5 D2 |
|
||
| **True runtime isolation** (container `--network=none --read-only --pids-limit`, nsjail) | [planned] — C6 is a static+ulimit scaffold only | Plan-07 C6 (V22) |
|
||
| Incident severity/kill-switch/runbook | [planned] | Plan-07 C7 (V23) |
|
||
| KMS key management (rotation, non-exportable) | [partial] — Vault Transit path implemented + validated live; not yet the default (local-key fallback), no HSM/short-lived IdP tokens | Plan-07 B3 |
|
||
| Reviewer approval workflow | [partial] — cryptographic **approval-identity** done (signed reviewer + role); live **IdP (OIDC/JWT)** + policy versioning/diff still planned | Plan-07 C4 (V20) |
|
||
| External append-only (WORM) audit | [partial] — hash-linked local ledger + rollback/tamper detection done; true WORM store (S3 Object Lock/QLDB) + trusted timestamp planned | Plan-07 C5 (V21) |
|
||
| Live CVE/OSV scanning wired in | [partial] — availability detected; local denylist authoritative offline | Plan-07 C2 follow-up |
|
||
|
||
## 4. Honest claim
|
||
|
||
Track A + Track C-MVP + Evidence Pack + H5 governance-hardening + H6 AgentOps-hardening
|
||
raise H4/H5/H6 from "PoC/demo (~3.0/5)" to **early internal-production hardening**, with
|
||
executable adversarial tests for every control (175 checks, 0 fail — last full run
|
||
2026-07-05; KMS validated live via Vault on 2026-07-04). Fair maturity score
|
||
(`00_SUBMISSION_PACKAGE/evidence/scoring-run-report.md`): per-harness **~80/100**,
|
||
**H5 76→80** (approval-identity + KMS live + WORM) and **H6 79→80** (live alert dispatch
|
||
+ provider-API reconciliation + hosted dashboard + window breaker), so the **lowest
|
||
harness is now 80** (H2/H4/H5/H6/H7 level) — CASAN **Level 4**, proven by attack. This is
|
||
**not** full production readiness: serious production still needs live IdP (OIDC/JWT), a
|
||
true WORM store (S3 Object Lock), KMS-by-default + HSM, true sandbox isolation,
|
||
multilingual detection, a deployed dashboard host + managed alert channel/on-call, and
|
||
real billing-API telemetry — the [partial]/[planned] rows above and in
|
||
`CASAN_PLAN_07_PRODUCTION_HARDENING.md`.
|