axstack 0.37.0 → 0.38.0

package/README.md

@@ -115,8 +115,8 @@ subject to explicit stop-after-publication or observation requests.
 See [Host operations](docs/host-operations.md#chat-run-watch-activation) for wake activation and cadence.
 See [Chat-run PR watch](docs/workflows.md#chat-run-pr-watch) for the lifecycle and quiet cadence.
 
-An optional native T3 review manager runs finite peer-review passes every 15
-minutes; the review automation never merges for you. Activation needs live
+An optional native T3 review manager runs finite peer-review passes hourly
+(`everyMs:3600000`); the review automation never merges for you. Activation needs live
 host validation. See [PR-manager setup and safety](docs/host-operations.md#optional-native-peer-review-automation).
 
 

package/docs/host-operations.md

@@ -52,7 +52,8 @@ Every verified own-PR publication arms or joins the driver's chat-run watch.
 Its bound T3 schedule uses the cadence in [Chat-run watch activation](#chat-run-watch-activation).
 See [Chat-run PR watch](workflows.md#chat-run-pr-watch) for authority, schedule identity and stop conditions.
 Missing schedule capability holds activation.
-The optional review manager uses an unbound 15-minute T3 schedule and requires
+The optional review manager uses an unbound hourly T3 schedule
+(`everyMs:3600000`) and requires
 its separate native canary before activation. Installed guidance does not prove
 live behavior. See [Review manager](../skills/axstack/references/automations.md).
 
@@ -172,8 +173,10 @@ with the PR; remove the exact unit and route after settlement.
 
 The optional native review manager uses the VPS T3 project `axstack-review-lane`
 on the existing host clone. Configure and read back the lane's `axstack-owner`
-binding, then create an unbound T3 schedule every 15 minutes. Each pass starts
-in a fresh finite worktree from `origin/main`, fetches first, and checks its
+binding, then create an unbound hourly T3 schedule (`everyMs:3600000`).
+The packaged prompt checks for a quiet exit before skill loading; full passes
+follow the lifecycle below. Each pass starts in a fresh finite worktree from
+`origin/main`, fetches first, and checks its
 binding. Continuity lives outside worktrees at
 `~/.local/share/axstack/runs/review-manager/progress.md`. Per-PR detached
 review checkouts come from existing host clones; a missing clone holds that job.
@@ -211,8 +214,9 @@ and [Single T3 server](#single-t3-server).
 
 ## Review automation
 
-The review manager uses one short packaged prompt that loads the current
-relative contract and invokes `axstack-review`. Bounded jobs publish ordinary
+The review manager uses one short packaged prompt that checks for a quiet exit,
+then loads the current relative contract and invokes `axstack-review` for a full pass.
+Bounded jobs publish ordinary
 exact-head review verdicts; peer PRs are merged by the user. Manual adopted-PR maintenance
 uses `axstack-watch` with local-SHA review before authorized publication.
 Exceptional security, permanent-on-chain, or architectural decisions remain actionable in GitHub or a durable user-owned conversation

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "axstack",
-  "version": "0.37.0",
+  "version": "0.38.0",
   "description": "Axstack installer and setup CLI: installs owned chat skills and role data, configures supported harness settings, and checks T3 Code capabilities.",
   "keywords": [
     "claude-code",

package/skills/axstack/references/automations.md

@@ -40,11 +40,16 @@ A missing, non-durable or unreadable continuity path holds admission.
 Follow the [Run record](run-record.md) and use the
 [Review-manager continuity template](run-record.md#review-manager-continuity-template):
 overwrite its four current-state sections and archive superseded history once
-beside it. A no-change pass appends at most one history line.
+beside it. A no-change full pass appends at most one history line.
+
+Follow the quiet check in the packaged [review-manager prompt](review-manager-prompt.md).
+The full pass records in Watermarks `Quiet gate:` as the sorted list and
+`Last full pass:` as UTC only as owner-of-record, then reads back the save.
+The sections below govern full passes; a quiet exit bypasses their teardown.
 
 Configure the lane thread via `t3_thread_configure` with the `axstack-owner`
 binding and verify its read-back. Then `schedule_task` uses the packaged prompt,
-`everyMs:900000`, `bindToCurrentThread:false`, and a stable `clientRequestId`.
+hourly with `everyMs:3600000`, `bindToCurrentThread:false`, and a stable `clientRequestId`.
 Record the schedule ID, project, lane binding and pass thread/run identities.
 Each pass compares its own `t3_thread_configuration` with the recorded binding
 after any owner-only instance update under [Provider bindings](t3-runtime.md#preflight-and-binding), before admission.

package/skills/axstack/references/lifecycle.md

@@ -131,7 +131,8 @@ differs from merged.
 ## Review automation health
 
 The native review manager uses fresh finite T3 pass threads from its dedicated
-lane project on a 15-minute T3 scheduled task. It admits actionable PR events
+lane project on an hourly T3 scheduled task (`everyMs:3600000`). It admits
+actionable PR events
 within measured host capacity; waiting PRs remain covered without reserving slots.
 Bounded PR jobs use driver-made detached checkouts and settle after descendants.
 Build no custom scheduler, state engine or legacy fallback. See

package/skills/axstack/references/review-manager-prompt.md

@@ -1,6 +1,44 @@
 # Review manager prompt
 
-At pass start, follow [Provider bindings](t3-runtime.md#preflight-and-binding) for own account selection.
+The account picker must run first, before the quiet check: from the installed
+`axstack` skill directory, run `bun scripts/pick-instance.js --provider <own provider> --json`.
+Find this pass's current instance from `t3_thread_configuration` in the picker
+JSON `instances` and inspect its `state`; defer reference loads until a full pass.
+An eligible current instance remains quiet-eligible even when `chosenInstanceId`
+names a different instance.
+Run the quiet check before invoking `axstack-review` or reading any `skills/**` file.
+
+Resolve self with `gh api user --jq .login`, then union two paginated `gh search prs`
+queries (`review-requested:@me` and `involves:@me`), each with `--limit 1000`,
+for open, non-draft PRs authored by someone else,
+producing a sorted (repo#number, headRefOid, updatedAt) list.
+Search omits head OIDs: obtain them with read-only `gh pr view --json headRefOid`.
+A result count at or above 1000 in either query makes coverage incomplete.
+Inventory every page of T3 lane threads and their run liveness, including recorded
+PR-job descendants, using exact lane identities.
+Read continuity at `~/.local/share/axstack/runs/review-manager/progress.md` and
+compare its Watermarks with the inventory:
+
+- Run the full pass when the picker JSON marks this pass's current instance `excluded`.
+- Run the full pass when the picker exits with code 1.
+- Run the full pass when the picker exits with code 2.
+- Run the full pass when the sorted list differs from `Quiet gate:`.
+- Run the full pass when any other lane thread is non-terminal.
+- Run the full pass when continuity or either watermark is missing, malformed or unreadable.
+- Run the full pass when `Last full pass:` is at least 4 hours old or in the future.
+- Run the full pass when any call fails or pagination, PR head OIDs or thread/run liveness is incomplete or unknown.
+- Run the full pass when continuity records any unserved, capacity-deferred, `INCOMPLETE` or re-eligible event, or any open hold with a resume condition beyond a user decision (such as capacity or remeasure).
+
+Holds awaiting only the user remain quiet-eligible: a user decision on the PR
+changes `updatedAt` and the 4 h full pass bounds reminders.
+
+If every busy condition is false, end the turn with no text, no skill load, no sweep and no continuity write.
+Otherwise continue with the full pass below; incomplete evidence retains the
+existing admission and cleanup holds.
+
+## Full pass
+
+Follow [Provider bindings](t3-runtime.md#preflight-and-binding) for own account selection.
 
 You are a fresh finite T3 review-manager pass in project `axstack-review-lane`
 on the VPS's existing `axatbhardwaj/axstack` clone. Fetch first; this unbound
@@ -17,5 +55,8 @@ Report retained worktree count and disable the schedule and hold past its
 authorized limit. Only the owner retires settled predecessors and saves and
 reads back continuity at
 `~/.local/share/axstack/runs/review-manager/progress.md` using
-`run-record.md#review-manager-continuity-template`; settle descendants and end
-the finite turn. Stay quiet when unchanged. Never check out a PR branch here.
+`run-record.md#review-manager-continuity-template`.
+The full pass records in Watermarks `Quiet gate:` as the sorted list and
+`Last full pass:` as UTC only as owner-of-record, then reads back the save.
+Settle descendants and end the finite turn. Stay quiet when unchanged.
+Never check out a PR branch here.

package/skills/axstack/references/role-roster.md

@@ -58,11 +58,16 @@
   `axstack-research-x`, `axstack-checker`, `axstack-research-code-sol`,
   `axstack-explore-execution-sol`, `axstack-auditor-sol`,
   `axstack-explore-codebase-flash`.
-  Always dispatch optional seats when configured and available. Only a
+  Always dispatch optional seats when configured and available, except research
+  branches, which follow [Research step 2](../../axstack-research/SKILL.md#procedure) sizing.
+  A research branch unneeded by an open claim is recorded `not needed (<reason>)`,
+  never as absent or a malfunction. Only a
   launch failure, trust/login prompt, or prompt block makes one absent:
   fence the attempt, record `absent (<reason>)`, name it once in the next
   read-back, and skip it without a relay or substitution. An uncertain
-  dispatch is reconciled, not called absent. A mixed fan-out still includes
-  at least one Codex and one Claude seat; otherwise hold that fan-out.
+  dispatch is reconciled, not called absent. A sized mixed research fan-out
+  requires dispatched branches spanning at least two different providers;
+  otherwise hold that fan-out. Every other mixed fan-out, including arena and
+  peer review, retains at least one Codex and one Claude seat; otherwise hold that fan-out.
   Intentional absences in codex-only and claude-only remain unchanged.
 - `axstack-debug-investigator-1..4` probe L1 briefs.

package/skills/axstack/references/routing.md

@@ -44,7 +44,8 @@ An exhausted account with no eligible sibling still holds; dispatched roles neve
 Follow [Provider bindings](t3-runtime.md#preflight-and-binding) for driver account re-selection at turn boundaries.
 [Model discipline](contracts.md#model-discipline) governs optional seats,
 auditor preflight, and required holds; [Role roster](role-roster.md) governs
-single-provider absence and mixed Codex+Claude fan-out.
+single-provider absence, mixed research with at least two different providers,
+and other mixed Codex+Claude fan-out (including arena and peer review).
 Bundled presets are setup inputs, not runtime readiness proof.
 
 Load the [Role roster](role-roster.md) for configured roles and authored-review pairings.

package/skills/axstack/references/run-record.md

@@ -159,6 +159,8 @@ in the adjacent history file, not below this template.
 - <PR or lane, reason, evidence, owner, resume condition; or none>
 
 ## Watermarks
+- Quiet gate: <sorted (repo#number, headRefOid, updatedAt) list; [] when empty>
+- Last full pass: <UTC timestamp of the last full pass>
 - <PR/event identity, exact head/base, last observed receipt; or none>
 
 ## Last pass

package/skills/axstack-research/SKILL.md

@@ -28,10 +28,20 @@ is part of research.
    gathering evidence only after the target and stopping boundary are explicit.
 
 2. **Fan out research:** A single factual lookup stays in the current chat.
-   Every other research run dispatches every configured research branch through
-   T3 `delegate_task`: requirements, code, and web (Sonnet high in mixed/claude-only;
-   Codex in codex-only), web-google (Gemini/Antigravity, with Google Search
-   built in), and X (Grok).
+   For other runs, dispatch through T3 `delegate_task` only branches whose
+   source class (requirements, code, execution, web, X) an open answer-changing claim needs.
+   Dispatch requirements only when scope is unclear.
+   Dispatch execution when a claim depends on runtime behavior.
+   Dispatch X only when a claim depends on X posts.
+   Record each skipped branch as `not needed (<reason>)`.
+   For each needed source class, retain two isolated branches where configured.
+   In mixed, each needed source class retains branches from different providers.
+   The code pair is `axstack-research-code` +
+   `axstack-research-code-sol`; the web pair is `axstack-research-web` +
+   `axstack-research-web-google`; the execution pair is `axstack-explore-execution` +
+   `axstack-explore-execution-sol`. Requirements, code, and web use Sonnet high
+   in mixed/claude-only and Codex in codex-only; web-google uses
+   Gemini/Antigravity with Google Search built in, and X uses Grok.
    Give each branch one owner, allow no cross-reading, and require a cited note
    with a URL and access date per claim; re-open sources and never trust a search
    summary. The driver reconciles agreements/disagreements per claim.
@@ -61,8 +71,10 @@ is part of research.
    never averaging findings. Record an intentionally absent pair and proceed
    with the base seat alone. The `-sol` pair is optional: fence a launch failure,
    trust/login prompt, or prompt block; record `absent (<reason>)`, and continue
-   with the base seat. In mixed fan-out,
-   retain a Codex and a Claude seat or hold the affected fan-out.
+   with the base seat. A sized mixed research fan-out requires dispatched
+   branches spanning at least two different providers; otherwise hold that fan-out.
+   Every other mixed fan-out, including arena and peer review, retains at least one
+   Codex and one Claude seat; otherwise hold that fan-out.
 
 3. **Gather primary source evidence.** Inspect the actual documentation, code,
    or tool output for every answer-changing claim. Apply the source standards