tangle-network · drewstone · Aug 12, 2026 · Aug 12, 2026
diff --git a/.agent/autopsies/2026-08-12-tangle-retained-canary.md b/.agent/autopsies/2026-08-12-tangle-retained-canary.md
@@ -0,0 +1,53 @@
+# Tangle cloud canary autopsy
+
+## Run
+
+The production checks used Braid commit `8f066cc` plus the retained-session worktree.
+They used OpenCode, `tangle-router/glm-5.2`, Runtime through `0.132.12`, and Sandbox SDK `0.21.1`.
+The checks used fresh bounded credentials that were revoked after each check.
+
+## Verified findings
+
+- The post-hardening retained canary failed before allocation in 4.695 seconds.
+- The published provider lacks exact lookup after an unacknowledged dispatch.
+- Active Sandboxes stayed at four.
+- Total Sandboxes stayed at 9,241.
+- Compute, GPU, and cost counters had zero measured change.
+- The exact Braid owner-tag query matched zero resources.
+- An earlier retained diagnostic reached an older sidecar and was rejected for its `runControlRef` field.
+- That earlier attempt moved total Sandboxes from 9,239 to 9,240 but returned active count to four.
+- A later ephemeral Braid canary timed out after 182.581 seconds with no environment identifier.
+- The Braid run remained `streaming` after Runtime emitted only `backend_start`.
+- A direct SDK create with the same route failed in 2.177 seconds.
+- The outer response was HTTP 400 with `CONFIG_ERROR`.
+- The nested response was HTTP 403 with `Service "sandbox" is not authorized for this endpoint`.
+- The same key authenticated Sandbox usage and Router model discovery.
+- Direct SDK account counts stayed at four active and 9,241 total Sandboxes.
+- Runtime `0.132.12` still treats the permanent error as retryable because its message contains `provision failed`.
+- Sandbox npm `latest` points at `0.21.0`, but Runtime `0.132.12` requires Sandbox `>=0.21.1 <0.22.0`.
+
+## Classification
+
+Retained mode is safely unavailable because the provider does not supply exact recovery lookup.
+Ephemeral cloud creation is currently blocked by internal platform authorization.
+Runtime then hides that terminal failure behind an excessive retry.
+The npm tag can also install an incompatible SDK version.
+These are three independent upstream failures.
+Braid created no resource during either post-hardening canary.
+
+## Action
+
+Issue `tangle-network/agent-dev-container#5249` tracks the earlier missing-turn response mismatch.
+Issue `tangle-network/agent-dev-container#5251` tracks the retained sidecar rollout.
+Issue `tangle-network/agent-dev-container#5277` tracks the current platform authorization failure.
+Issue `tangle-network/agent-dev-container#5278` tracks the incorrect npm tag.
+Issue `tangle-network/agent-runtime#808` tracks permanent-error retry classification.
+Braid keeps retained mode fail-closed and defaults new Sandbox connections to ephemeral mode.
+The production canary must pass before the bounded durability cohort starts.
+
+## Artifacts
+
+- `artifacts/verification/live/tangle-sandbox-braid-retained-canary-production-20260812.json`
+- `artifacts/verification/live/tangle-sandbox-braid-retained-safe-preflight-production-20260812.json`
+- `artifacts/verification/live/tangle-sandbox-braid-retained-post-hardening-production-20260812.json`
+- `artifacts/verification/live/tangle-sandbox-braid-execution-canary-post-hardening-production-20260812.json`
diff --git a/.agent/critical-audit/2026-08-12T17-22-32Z/findings.jsonl b/.agent/critical-audit/2026-08-12T17-22-32Z/findings.jsonl
@@ -0,0 +1,7 @@
+{"id":1,"severity":"HIGH","file":"scripts/live-required/tangle-sandbox-braid-stress.mjs","line":594,"defect":"Public cloud proof serialized raw customer and billing-owner identifiers.","scenario":"An authenticated live proof writes its account observation, then a committed public artifact discloses two real account identifiers.","status":"measured · resolved","evidence":"scripts/live-required/tangle-sandbox-braid-stress.mjs:588-594; actual artifact scan → 5/5 JSON files contain zero customerId or billingOwnerId keys","fix":"Compare raw identity only in memory and emit one SHA-256 identity digest.","verification":"public artifact credential-pattern scan → 5/5 pass; raw-account-key scan → 5/5 pass; pnpm check → 681 pass, 2 skip, 0 fail","costIfShipped":"2 account identifiers disclosed per published proof artifact","savedIfFixed":"2 of 2 account identifiers removed per artifact"}
+{"id":2,"severity":"HIGH","file":"src/adapters/runtime/retained-execution.ts","line":75,"defect":"Detach during an unresolved cloud start reopened the event reader after start completed.","scenario":"A user detaches while start is pending, the provider returns a handle, then Braid reads a run the user already detached and can duplicate reconnect delivery.","status":"measured · resolved","evidence":"src/adapters/runtime/retained-execution.ts:66-76; test/retained-execution-lifecycle.test.ts:206-245","fix":"Persist detached state before aborting and stop streamTurn after the start promise resolves.","verification":"pnpm check → in-flight detach test passed; 681 pass, 2 skip, 0 fail","costIfShipped":"1 unintended reader per in-flight detached run","savedIfFixed":"1 of 1 unintended reader prevented per run"}
+{"id":3,"severity":"MEDIUM","file":"src/adapters/runtime/tangle-sandbox-retention.ts","line":53,"defect":"The retained client wrapper dropped identity, usage, and subscription methods.","scenario":"A retained connection wraps a Sandbox client, then Braid cannot observe account identity or spend and reports three available provider surfaces as unknown.","status":"measured · resolved","evidence":"src/adapters/runtime/tangle-sandbox-retention.ts:53-55,89-91; test/tangle-retained-lifecycle.test.ts:142-173","fix":"Bind and preserve every optional account observation method on the wrapper.","verification":"pnpm check → retained account-observation test passed; 681 pass, 2 skip, 0 fail","costIfShipped":"3 account observation methods lost per wrapped client","savedIfFixed":"3 of 3 methods preserved per client"}
+{"id":4,"severity":"MEDIUM","file":"scripts/live-required/tangle-sandbox-braid-stress.mjs","line":926,"defect":"A lost shutdown acknowledgement skipped process-tree cleanup proof.","scenario":"Shutdown RPC fails after delivery, then the stress runner throws before checking child exit and cannot prove that descendants stopped.","status":"measured · resolved","evidence":"scripts/live-required/tangle-sandbox-braid-stress.mjs:926-947; test/tangle-sandbox-braid-stress-runtime.test.mjs:123-157","fix":"Always close and verify the process tree, then attach that proof to the original shutdown error.","verification":"pnpm check → shutdown-failure cleanup test passed; live-required self-test 26/26 passed","costIfShipped":"1 process tree unverified per shutdown acknowledgement failure","savedIfFixed":"1 of 1 process tree verified per failure"}
+{"id":5,"severity":"MEDIUM","file":"src/app/run-replay.ts","line":53,"defect":"Recovery replaced a typed dispatch failure when exact provider status was unavailable.","scenario":"Cloud provisioning returns CLOUD_PROVISION_REJECTED and status lookup is absent, then Braid shows a generic unknown-state message instead of the safe typed cause.","status":"measured · resolved","evidence":"src/app/run-replay.ts:53-55,76-82,104-110; test/w8-runs.test.ts:251-270","fix":"Carry the sanitized prior failure through reconnect and null-status reconciliation.","verification":"pnpm check → typed diagnostic test passed; 681 pass, 2 skip, 0 fail","costIfShipped":"1 actionable diagnostic lost per affected failed dispatch","savedIfFixed":"1 of 1 diagnostic retained per affected dispatch"}
+{"id":6,"severity":"MEDIUM","file":"src/domain/invariants-profile.ts","line":140,"defect":"Non-Sandbox connections accepted Sandbox lifecycle fields.","scenario":"A CLI Bridge connection includes retained lifecycle and idle TTL, then validation accepts provider options that its adapter cannot implement.","status":"measured · resolved","evidence":"src/domain/invariants-profile.ts:140-145; test/domain-invariants.test.ts:199-206","fix":"Reject lifecycle and idle TTL unless the connection kind is tangle-sandbox.","verification":"pnpm check → domain invariant test passed; 681 pass, 2 skip, 0 fail","costIfShipped":"2 unsupported fields accepted per malformed connection","savedIfFixed":"2 of 2 unsupported fields rejected per connection"}
+{"id":7,"severity":"LOW","file":"docs/05-profiles-and-connections.md","line":275,"defect":"Documentation claimed an operator-selectable retained control that the connection editor does not expose.","scenario":"A terminal user follows the setup documentation, then no retained lifecycle control exists in the five-field editor.","status":"measured · resolved","evidence":"src/views/tui/connection-metadata-editor-model.ts:15-21; docs/05-profiles-and-connections.md:273-277","fix":"Describe retained lifecycle as a connection configuration option until the provider path and editor control are available.","verification":"rg connection editor fields → kind, name, endpoint, region, account only; pnpm format:check passed","costIfShipped":"1 unavailable UI action claimed per reader path","savedIfFixed":"1 of 1 unavailable action claims removed"}
diff --git a/.agent/critical-audit/2026-08-12T17-22-32Z/manifest.json b/.agent/critical-audit/2026-08-12T17-22-32Z/manifest.json
@@ -0,0 +1,26 @@
+{
+  "schemaVersion": 1,
+  "scope": "Braid retained Tangle Sandbox lifecycle, stress proof, and public evidence",
+  "files": 65,
+  "base": "8f066cc20c750579015bf47589d3d742d6bb0bc5",
+  "head": "8f066cc20c750579015bf47589d3d742d6bb0bc5",
+  "workingTree": true,
+  "diffSha256": "2389bc43d72a13cf9bbd348d57d8f7a9955b2673d844a3953d5856f3c4c9b067",
+  "projectType": "Node.js ESM package",
+  "reviewers": [
+    "A: retained lifecycle correctness, identity, cleanup, credentials, and races",
+    "B: runtime boundaries, module ownership, replay, and failure handling",
+    "C: callsites, documentation accuracy, dependency hygiene, and real-system proof"
+  ],
+  "mode": "current working-tree review with serial local checks",
+  "verdict": "APPROVE",
+  "counts": {
+    "critical": 0,
+    "high": 2,
+    "medium": 4,
+    "low": 1,
+    "resolved": 7,
+    "unresolved": 0,
+    "dropped": 0
+  }
+}
diff --git a/.agent/critical-audit/2026-08-12T17-22-32Z/summary.md b/.agent/critical-audit/2026-08-12T17-22-32Z/summary.md
@@ -0,0 +1,41 @@
+# Audit: Braid retained Tangle Sandbox lifecycle and proof — 8f066cc20c750579015bf47589d3d742d6bb0bc5..8f066cc20c750579015bf47589d3d742d6bb0bc5 — n=65 files, 7 findings
+
+**Verdict:** APPROVE — 7 of 7 measured defects resolved · 0 CRITICAL / 2 HIGH / 4 MEDIUM / 1 LOW
+**Worst:** #1 `scripts/live-required/tangle-sandbox-braid-stress.mjs:594` — public proof disclosed two account identifiers · cost if shipped 2 identifiers per artifact
+**Next:** `/verify`
+
+## Scope
+
+| Field | Value |
+|---|---|
+| Files | n=65 via current working-tree diff |
+| Base..head | `8f066cc20c750579015bf47589d3d742d6bb0bc5..8f066cc20c750579015bf47589d3d742d6bb0bc5` |
+| Project type | Node.js ESM package |
+| Reviewers | A,B,C · serial |
+| Not inspected | Provider implementation outside Braid; live retained execution remains blocked by tracked upstream issues. |
+
+## Findings — 7 of 7, ranked
+
+| # | Sev | file:line | Defect | Failure scenario (input/state → wrong result) | Status | Evidence | Fix | Verification | Cost if shipped | Saved if fixed |
+|---:|---|---|---|---|---|---|---|---|---:|---:|
+| 1 | HIGH | `scripts/live-required/tangle-sandbox-braid-stress.mjs:594` | Public proof emitted raw account identity. | Authenticated proof → two identifiers published. | measured · resolved | actual artifacts: 5/5 have zero raw identity keys | Emit one digest. | public scans 5/5 pass | 2 identifiers/artifact | 2/2 removed |
+| 2 | HIGH | `src/adapters/runtime/retained-execution.ts:75` | Pending detach reopened the event reader. | Detach before start resolves → detached run streams locally. | measured · resolved | `test/retained-execution-lifecycle.test.ts:206` | Stop after start when detached. | `pnpm check` test passed | 1 reader/run | 1/1 prevented |
+| 3 | MEDIUM | `src/adapters/runtime/tangle-sandbox-retention.ts:53` | Wrapper dropped account methods. | Wrap client → identity, usage, and subscription become unknown. | measured · resolved | `test/tangle-retained-lifecycle.test.ts:142` | Preserve bound methods. | `pnpm check` test passed | 3 methods/client | 3/3 preserved |
+| 4 | MEDIUM | `scripts/live-required/tangle-sandbox-braid-stress.mjs:926` | Shutdown error hid cleanup proof. | Lost acknowledgement → descendants remain unverified. | measured · resolved | `test/tangle-sandbox-braid-stress-runtime.test.mjs:123` | Verify close before rethrow. | live-required 26/26 | 1 tree/failure | 1/1 verified |
+| 5 | MEDIUM | `src/app/run-replay.ts:53` | Recovery dropped the typed cause. | Failed dispatch plus no status → generic error. | measured · resolved | `test/w8-runs.test.ts:251` | Carry safe prior detail. | `pnpm check` test passed | 1 diagnostic/run | 1/1 retained |
+| 6 | MEDIUM | `src/domain/invariants-profile.ts:140` | Other connections accepted Sandbox lifecycle fields. | CLI Bridge plus retained TTL → invalid configuration accepted. | measured · resolved | `test/domain-invariants.test.ts:199` | Restrict fields by kind. | `pnpm check` test passed | 2 fields/config | 2/2 rejected |
+| 7 | LOW | `docs/05-profiles-and-connections.md:275` | Docs claimed a missing editor action. | User follows setup text → no retained control exists. | measured · resolved | `src/views/tui/connection-metadata-editor-model.ts:15` | State configuration boundary. | editor field inspection | 1 claim/reader | 1/1 removed |
+
+0 dropped (no reproducible failure scenario / no actionable fix).
+
+## Assumptions & unverified
+
+| Assumption | Finding it would flip | Check that settles it |
+|---|---|---|
+| The provider will add exact retained-run lookup without changing the six-field control reference. | The current adapter contract could require a new shared field. | Resolve upstream runtime issue 800 and provider issue 146, then rerun the retained stress command. |
+| The current Sandbox authorization regression is external to Braid. | A Braid credential-routing defect could be present. | Resolve ADC issue 5277, then rerun both direct SDK and Braid canaries with one new key. |
+
+## Self-gate
+
+9/9 passed — failed: none.
+1 verdict = decision + 1 number · 2 every finding has file:line · 3 concrete failure scenario · 4 status label · 5 evidence is a pointer · 6 cost both sides · 7 fix + verification per row · 8 zero adjectives standing in for counts · 9 148 words ≤600 outside tables.
diff --git a/.agent/skill-runs.jsonl b/.agent/skill-runs.jsonl
@@ -41,3 +41,13 @@
 {"skill":"/harden","ts":"2026-08-12T03:42:22Z","project":"braid-transcript-first-wt","target":"provider-neutral retained execution lifecycle, exact provider identity, recovery, cancellation, and event binding n=11 files","operatorPrompt":"","durationMin":null,"verdict":"APPROVE","dispatchedTo":"/critical-audit","operatorOverride":null,"transcriptPath":null,"traceDir":null}
 {"skill":"/critical-audit","ts":"2026-08-12T03:42:23Z","project":"braid-transcript-first-wt","target":"provider-neutral retained execution lifecycle and provider-owned identity binding n=11 files","operatorPrompt":"","durationMin":null,"verdict":"APPROVE","dispatchedTo":"/stop","operatorOverride":null,"transcriptPath":null,"traceDir":null}
 {"skill":"/converge","ts":"2026-08-12T12:53:14Z","project":"braid-transcript-first-wt","target":"Braid full release check performance failure","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/review-to-green","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/sandbox-sdk-integration","ts":"2026-08-12T14:30:51Z","project":"braid-retained-v2-wt","target":"Braid retained Tangle sandbox execution through the published SDK and provider","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/ground-truth","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/ground-truth","ts":"2026-08-12T14:30:51Z","project":"braid-retained-v2-wt","target":"production retained cloud create, dispatch, reconnect, cleanup, usage, and latency path n=1","operatorPrompt":"","durationMin":null,"verdict":"EXTERNAL_BLOCKER_CONFIRMED","dispatchedTo":"/diagnose","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/diagnose","ts":"2026-08-12T14:30:51Z","project":"braid-retained-v2-wt","target":"retained Tangle canary failure n=1","operatorPrompt":"","durationMin":null,"verdict":"ROOT_CAUSE_CONFIRMED","dispatchedTo":"/autopsy","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/autopsy","ts":"2026-08-12T14:30:51Z","project":"braid-retained-v2-wt","target":"production retained canary and direct start traces n=2","operatorPrompt":"","durationMin":null,"verdict":"INFRA_BUG","dispatchedTo":"/verify","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/bp-ops","ts":"2026-08-12T14:30:51Z","project":"braid-retained-v2-wt","target":"Sandbox production health, deploy history, and resource cleanup","operatorPrompt":"","durationMin":null,"verdict":"DEPLOY_BLOCKED","dispatchedTo":"/adc-infra-triage","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/adc-infra-triage","ts":"2026-08-12T14:30:52Z","project":"braid-retained-v2-wt","target":"production and staging ADC revisions versus exact retained-run revision","operatorPrompt":"","durationMin":null,"verdict":"ROOT_CAUSE_CONFIRMED","dispatchedTo":"/verify","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/report","ts":"2026-08-12T14:30:52Z","project":"braid-retained-v2-wt","target":"Braid retained Tangle production status and stress evidence","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/stop","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/harden","ts":"2026-08-12T17:28:15Z","project":"braid-retained-v2-wt","target":"Braid retained Tangle lifecycle, public proof redaction, replay races, and cleanup n=65 files","operatorPrompt":"","durationMin":null,"verdict":"APPROVE","dispatchedTo":"/critical-audit","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/critical-audit","ts":"2026-08-12T17:28:16Z","project":"braid-retained-v2-wt","target":"Braid retained Tangle lifecycle, stress proof, and public evidence n=65 files","operatorPrompt":"","durationMin":null,"verdict":"APPROVE","dispatchedTo":"/verify","operatorOverride":null,"transcriptPath":null,"traceDir":null}
+{"skill":"/verify","ts":"2026-08-12T17:28:16Z","project":"braid-retained-v2-wt","target":"Braid retained Tangle source, 683 tests, packed terminal, cloud proof contracts, and CLI Bridge matrix","operatorPrompt":"","durationMin":null,"verdict":"PASS","dispatchedTo":"/ship","operatorOverride":null,"transcriptPath":null,"traceDir":null}