jefahnierocks · verlyn13 · Aug 10, 2026 · Aug 8, 2026 · Aug 8, 2026 · Aug 8, 2026
@@ -9,11 +9,12 @@ and the user-level MCP baseline.
 
 Read these for current state:
 
+- [`docs/restart-handoff-2026-08-08.md`](./docs/restart-handoff-2026-08-08.md) — dated restart snapshot for a fresh agent: verified Git and host state, updater evidence, remaining work, scope boundaries, and first commands. It routes to the authorities below and does not replace them
 - [`docs/ai-tool-config-ownership.md`](./docs/ai-tool-config-ownership.md) — **the ownership boundary across every AI tool**: which of the four mechanisms deploys each path, the global-user config tree, the project-scope model, the established managed/root tier, the git-identity registration gap, and the leftover-workflow register. Read before editing any agent config file
 - [`docs/project-conventions.md`](./docs/project-conventions.md) — compatibility guide for downstream projects (link this from a project's own `AGENTS.md`)
 - [`docs/secrets.md`](./docs/secrets.md) — secret-handling policy (1Password)
 - [`docs/1password-service-account.md`](./docs/1password-service-account.md) — **live** operator runbook for the `mcp-gui-readonly` service account: the non-interactive `op` auth plane that GUI/headless MCP wrappers (brave-search, firecrawl, runpod) are coded to use. **Provisioning state 2026-07-27: LIVE** — service account created (read-only on `Dev`), keychain item present and intact, `ng-doctor mcp` 6/6. Read the token-storage box before rotating: `security add-generic-password -w` truncates at 128 bytes and cannot store an SA token; use `scripts/store-op-sa-token.sh`. Indexed here directly because it was previously reachable only through the two draft platform docs below, which this file labels "not live provisioning policy" — the runbook that would have caught the failure was two hops behind a "not live" sign
-- [`docs/secret-records.md`](./docs/secret-records.md) — credential-record register (identifier, owner, consumer, scope, rotation, stop rules; no secret values). The SA token's record is tracked here with status `planned`
+- [`docs/secret-records.md`](./docs/secret-records.md) — credential-record register (identifier, owner, consumer, scope, rotation, stop rules; no secret values). The SA token's current record status is `active`; the dated 2026-05-28 transition entry remains historical evidence of its earlier `planned` state
 - [`docs/infisical-cli.md`](./docs/infisical-cli.md) — Infisical CLI baseline for project runtime secret workflows; Homebrew core install/update path and syntax guardrails
 - [`docs/homebrew-tap-trust.md`](./docs/homebrew-tap-trust.md) — Homebrew 6 third-party tap trust boundary; item-level trust policy, operator review workflow, and fail-closed updater behavior
 - [`docs/1password-platform-architecture.md`](./docs/1password-platform-architecture.md) — draft target architecture for 1Password developer-platform adoption; not live provisioning policy
@@ -209,12 +210,9 @@ promote it in `collect_notices` instead.
 `ProgramArguments` point into this repo. It has run every 60s since 2026-04-24,
 writing daily JSONL to `~/.local/state/system-config/mcp-usage-collector/` and
 logs to `~/Library/Logs/system-config/`. **There is no retention policy and the
-state directory has grown to 6.3 GB across 95 daily JSONL files** (`du -shk`,
-2026-07-27; this figure supersedes the "roughly 13 GB" previously stated here and
-the "roughly 6.7 GB" previously stated in the ownership map — the same directory
-was carrying two different numbers). It grows continuously; treat it as a dated
-reading, not a constant. The plist itself has no chezmoi source and no drift
-check. See
+state directory has grown to 7.3 GB across 107 daily JSONL files** (`du -sk` =
+7,603,080 KiB, 2026-08-08). It grows continuously; treat this as a dated reading,
+not a constant. The plist itself has no chezmoi source and no drift check. See
 [`docs/ai-tool-config-ownership.md`](./docs/ai-tool-config-ownership.md) §1.1.
 
 ## Source of truth

@@ -3,8 +3,8 @@ title: AI Tool Configuration Ownership Map
 category: reference
 component: ai_tool_config_ownership
 status: active
-version: 1.3.1
-last_updated: 2026-08-01
+version: 1.3.2
+last_updated: 2026-08-08
 tags: [claude, codex, cursor, devin, copilot, gemini, aider, mcp, chezmoi, ownership, managed-settings, git-identity]
 priority: high
 ---
@@ -96,11 +96,11 @@ currently overstate.
 A fifth mechanism exists and is undocumented elsewhere in this repo:
 **`scripts/mcp-usage-collector.sh`** writes
 `~/Library/LaunchAgents/com.jefahnierocks.mcp-usage-collector.plist`, whose
-`ProgramArguments` point into this repo, and has produced **6.3 GB** of MCP
-telemetry across **95 daily JSONL files** under
+`ProgramArguments` point into this repo, and has produced **7.3 GB** of MCP
+telemetry across **107 daily JSONL files** under
 `~/.local/state/system-config/mcp-usage-collector/` since 2026-04-24 —
-re-measured by `du -shk` on 2026-07-27 **[V]**. It is repo-deployed by mechanism,
-unowned by documentation. See §7.
+re-measured as **7.3 GB** (`du -sk` = 7,603,080 KiB) on 2026-08-08
+**[V]**. It is repo-deployed by mechanism, unowned by documentation. See §7.
 
 > Earlier revisions of this document carried **two different figures for the same
 > directory** — "roughly 6.7 GB" here and "~13 GB" in §2.3 — and `AGENTS.md`
@@ -229,7 +229,7 @@ tooling prunes **[A]**:
 | --- | --- | --- |
 | `~/.codex/` (`logs_2.sqlite` 1.0 GB + WAL 386 MB, 1741 session transcripts) | ~7.3 GB | runtime |
 | `~/Library/Application Support/Cursor/User/globalStorage/state.vscdb` (+ backup) | ~4.1 GB | runtime, secret-bearing |
-| `~/.local/state/system-config/mcp-usage-collector/` (95 daily JSONL) | 6.3 GB **[V, 2026-07-27]** | runtime, repo-produced |
+| `~/.local/state/system-config/mcp-usage-collector/` (107 daily JSONL) | 7.3 GB **[V, 2026-08-08]** | runtime, repo-produced |
 | `~/.local/share/claude/versions/` (4 retained binaries) | ~1.0 GB | runtime |
 | `~/Library/Caches/claude-cli-nodejs/` | 346 MB | runtime |
 

@@ -0,0 +1,172 @@
+---
+title: "Restart Handoff — 2026-08-08"
+category: handoff
+component: workstation
+status: active
+version: 1.0.0
+last_updated: 2026-08-08
+tags: [restart, handoff, system-update, git, identity, mcp, vscode]
+priority: high
+---
+
+# Restart Handoff — 2026-08-08
+
+This is a dated, evidence-backed orientation snapshot for the next agent after a
+workstation restart. It records direct local and GitHub observations from
+2026-08-08 in America/Anchorage. It does not supersede `AGENTS.md` or any
+task-specific authority linked there. Re-run the commands below before acting on
+mutable state.
+
+## Start here
+
+1. Read [`../AGENTS.md`](../AGENTS.md) in full.
+2. Read this handoff and establish exact Git/PR state with the commands below.
+3. Follow the task-specific authority from `AGENTS.md`; do not treat this dated
+   snapshot as permission to modify another repository or operator-owned state.
+
+```bash
+pwd
+git status -sb
+git log -3 --show-signature --oneline
+git rev-list --left-right --count HEAD...@{upstream}
+gh pr view 61 --json number,title,state,isDraft,url,headRefName,baseRefName,statusCheckRollup
+```
+
+## Durable Git and publication state
+
+Verified at 2026-08-08 09:31 AKDT, before this handoff document was added:
+
+- Work branch: `agent/system-update-restart-readiness`.
+- Draft PR: [#61](https://github.com/jefahnierocks/system-config/pull/61),
+  targeting `main` from that branch.
+- The branch was clean and matched its upstream (`0 0`).
+- It was two signed commits ahead of `origin/main`:
+  - `1ff05e6` — `fix(system-update): report and resolve updates truthfully`
+  - `33852e4` — `docs: refresh verified host state`
+- `origin/main` was `bdec977`; PR #61 was open and its validation jobs had
+  started. GitHub status is mutable, so re-read it rather than assuming the PR
+  was merged or its checks remained green.
+
+This handoff and its `AGENTS.md` index entry follow those two commits on the same
+branch. The closeout gate is a clean worktree, `HEAD...@{upstream}` equal to
+`0 0`, valid signatures, and a successful PR validation check.
+
+## Updater incident and verified repair
+
+Full update run `20260808-080519` completed every step except stable npm globals.
+The stable lane passed exact `wrangler@4.120.0` with `--min-release-age 2`; that
+release was too new, so npm returned `ETARGET` rather than resolving backward.
+
+The repair keeps the exact inventory target for logs and comparison, but uses
+`<package>@latest` as the install spec while the stable cooldown is active. The
+install-script identity gate remains strict and owner-scoped. The AI channel
+lane is unchanged. The authoritative incident and implementation record is
+[`system-update-status-and-resilience-plan.md`](./system-update-status-and-resilience-plan.md)
+§§23–24.
+
+Verified evidence:
+
+- Focused updater suite: 156 passed, 0 failed.
+- Targeted live update `20260808-092739`: zero failures; Wrangler advanced from
+  `4.118.0` to eligible `4.119.0`; requested `4.120.0` remained held and was
+  disclosed by `npm-global-resolved-drift:`.
+- Installed readback: `wrangler --version` and `npm ls -g --depth=0 wrangler`
+  both reported `4.119.0`.
+- Full read-only check `20260808-093230`: all steps successful in 36 seconds;
+  Homebrew current; all three AI channels current with exact platform binaries;
+  Claude Code and gh extensions current.
+- The check inventory still reported Wrangler `4.120.0` and mise-managed uv
+  `0.12.3` as newer. This is inventory, not proof of a failed update. Re-run
+  `system-update --check` before deciding either item still needs work.
+- Full `scripts/validate-repo.sh`: PASS.
+- `chezmoi apply --dry-run --force`: exit 0.
+
+Relevant logs:
+
+```text
+~/Library/Logs/system-update/run-20260808-080519.log
+~/Library/Logs/system-update/run-20260808-092739.log
+~/Library/Logs/system-update/run-20260808-093230.log
+```
+
+## Verified host state
+
+- `ng-doctor` before commit closeout reported 71 passed, 3 failed, 20 skipped.
+  One failure was this repo's then-dirty worktree and should disappear when the
+  closeout gate is clean. The two identity failures below are external scope.
+- The GitHub MCP disable marker exists and containment passes; no managed host
+  exposes GitHub MCP while it exists.
+- The GUI/headless MCP service-account keychain item is present; all six MCP
+  doctor checks passed. The current secret-record status is `active`.
+- The MCP usage collector LaunchAgent is registered, interval-driven, and last
+  exited 0. Its state directory measured 7,603,080 KiB (7.3 GB) across 107 daily
+  JSONL files. No retention policy or chezmoi-owned plist source exists.
+- No updater process or `system-update.lock` was active during the initial
+  restart audit. Recheck after restart rather than relying on that dated fact.
+
+## Remaining work with current evidence
+
+### 1. External Git identity drift — do not repair from this repo
+
+`ng-doctor identity` reports 7 passed and 2 failed. The failures overlap across
+three real clones:
+
+| Clone path | Verified live fact |
+| --- | --- |
+| `/Users/verlyn13/Organizations/the-nash-group-github-profile` | Origin uses `github.com`, while the registry expects `github.com-nash-group`; local status already had 3 entries. The doctor prints this same clone as `parent-org/.github` in the remote check and `The-Nash-Group/.github` in the config-drift check. |
+| `/Users/verlyn13/Organizations/happy-patterns/apps/happy-patterns-org.github.io` | Clean worktree; origin uses `github.com`, while the registry expects `github.com-happy-patterns`; email/signing/host configuration also differs from the registered lane. |
+| `/Users/verlyn13/Organizations/happy-patterns/apps/scopecam` | Clean worktree on `development`; origin uses `github.com`, while the registry expects `github.com-happy-patterns`; email/signing/host configuration also differs from the registered lane. |
+
+The registry gates themselves pass: schema version 1, internal visibility, and
+zero unresolved repositories. These clone changes belong to their repositories
+and identity owner. Preserve the Nash clone's existing local changes.
+
+### 2. MCP telemetry retention and ownership
+
+The collector grows continuously and has no retention policy. Treat deletion,
+compression, LaunchAgent ownership, and migration as unapproved design work
+until the operator selects a policy. The ownership authority and full leftover
+register are [`ai-tool-config-ownership.md`](./ai-tool-config-ownership.md)
+§§1.1, 7, and 8.
+
+### 3. Operator-gated editor and posture work
+
+`ng-doctor` skipped 20 checks rather than proving them complete. Current examples
+include the staged VS Code profiles, 12 extension retirement candidates, and
+the phased macOS/network posture checks. Use
+[`vscode-configuration-plan.md`](./vscode-configuration-plan.md) and
+[`security-hardening-implementation-plan.md`](./security-hardening-implementation-plan.md)
+as planning authorities; re-verify live state and obtain the named operator
+decisions before mutation.
+
+## Do not infer
+
+- A draft PR is not merged authority.
+- A successful targeted npm update is not a substitute for re-reading the latest
+  full check before a future update.
+- An `outdated` inventory entry under a release-age policy is not automatically
+  a defect or permission to bypass the hold.
+- Doctor `skipped` means unverified or deferred, not passed.
+- External clone drift is evidence, not authorization for cross-repo edits.
+- Never expose secret values while validating the MCP service-account plane.
+
+## Fresh-agent restart commands
+
+These are read-only except the validation suite's temporary fixtures:
+
+```bash
+cd /Users/verlyn13/Organizations/jefahnierocks/system-config
+git status -sb
+git fetch --prune origin
+git rev-list --left-right --count HEAD...@{upstream}
+git log -3 --show-signature --oneline
+gh pr view 61 --json state,isDraft,url,statusCheckRollup
+system-update --check
+ng-doctor
+scripts/validate-repo.sh
+chezmoi apply --dry-run --force
+```
+
+Do not run `chezmoi apply`, merge PR #61, modify the three external clones, or
+delete telemetry solely because this handoff exists. Each requires its normal
+authority and a current readback.
@@ -3,8 +3,8 @@ title: "System-Update — Resilience & Status Plan"
 category: reference
 component: system_update
 status: active
-version: 1.13.0
-last_updated: 2026-07-29
+version: 1.15.0
+last_updated: 2026-08-08
 tags: [system-update, brand, terminal-style, jr-style, resilience, network, status, ux, npm, mise, concerns, homebrew, 1password, architecture, android-studio]
 priority: high
 ---
@@ -1481,6 +1481,48 @@ CLI invoked daily, "runtime" means every invocation. It also misses long-dwell
 compromise — event-stream hid for ~72 days — and a 2-day cooldown delays this
 host's own `npm audit fix` by 2 days.
 
+## 23. 2026-08-05 AI-channel check-summary truthfulness
+
+Read-only run `20260805-083114` resolved all three configured dist-tags
+successfully and rendered `available 3`. That count was not an update inventory:
+Gemini's installed version equalled its target, while Codex and Copilot were the
+only mismatches. The summary was counting `ai-cli-channel-ok:`, whose contract is
+"the registry lookup and channel health checks succeeded," not "an update is
+available."
+
+Check mode now compares each resolved target with the package manifest under the
+configured npm prefix and emits one of two additive markers:
+
+- `ai-cli-channel-current:` — installed and target versions are equal.
+- `ai-cli-channel-outdated:` — installed is absent or differs from the target.
+
+The result row derives its update count only from the second marker and reports
+the current count separately. Update mode retains `ai-cli-channel-ok:` as its
+successful-install count, so the correction does not reinterpret update runs or
+change channel installation behavior. The test harness reproduces the incident's
+one-current/two-outdated shape and requires `2 outdated, current 1`.
+
+## 24. 2026-08-08 stable-global cooldown resolution
+
+Update run `20260808-080519` asked for `wrangler@4.120.0` under the stable
+global lane's two-day release-age policy. That release was published on
+2026-08-07 and was inside the cooldown, so npm rejected the exact version with
+`ETARGET`: an exact ineligible target cannot resolve backward. Wrangler remained
+at `4.118.0` and the run ended with one failure.
+
+A read-only reproduction separated the policy from the target shape:
+
+- exact `wrangler@4.120.0` plus `--min-release-age 2` reproduced `ETARGET`;
+- `wrangler@latest` with the same gate resolved eligible `4.119.0`, published
+  2026-08-05, and completed its dry run successfully.
+
+The stable update step now retains the exact `npm outdated` target for log and
+post-install comparison, but passes `<package>@latest` to npm whenever a positive
+cooldown is active. npm can therefore select the newest eligible release, while
+`npm-global-resolved-drift:` still discloses that the inventory requested a newer
+version. With no cooldown, the exact inventory target remains the install spec.
+The AI channel lane is unchanged.
+
 ## Related
 
 - [`docs/brand/terminal-style.md`](./brand/terminal-style.md) — the four-tier brand spec (authority)