refactor(gate): cut over to direct author checks

.agents/roles/standards-enforcer.md

@@ -3,7 +3,7 @@ title: Standards Enforcer Role
 status: active
 owner: eng
 canonical: true
-last_verified: 2026-07-22
+last_verified: 2026-09-02
 doc_type: role
 scope: repo-wide
 review_interval_days: 90
@@ -18,7 +18,9 @@ Check changed files against repo instructions and context standards before PR re
 
 1. Read `docs/context-standards.md`.
 2. Identify applicable `AGENTS.md` files from repo root down to each changed file.
-3. Identify applicable checklists from `docs/pr-checklists/` and `pnpm agent:quality-gate --dry-run`.
+3. Identify applicable checklists from `docs/pr-checklists/`, scoped
+   instructions, and the author-check table in step 3 of the
+   [operating card](../../docs/notes/pr-operating-card.md).
 4. Run `pnpm agent:context-check`.
 5. Review for canonical/non-canonical confusion, missing scoped instructions, duplicated instructions, missing metadata, and root policy that belongs in nested AGENTS.
 

.agents/roles/verifier.md

@@ -3,7 +3,7 @@ title: Verifier Role
 status: active
 owner: eng
 canonical: true
-last_verified: 2026-08-26
+last_verified: 2026-09-02
 doc_type: role
 scope: repo-wide
 review_interval_days: 90
@@ -16,25 +16,16 @@ Run diff-scoped verification and report only actionable results.
 
 ## Procedure
 
-1. Inspect the branch diff against `origin/main`.
-2. Run `pnpm agent:quality-gate --dry-run` and confirm mapped commands/checklists match changed surfaces.
-3. Unless the requester asked for dry verification only, run the gate as a
-   background task. A local setup runs `pnpm agent:quality-gate --run` against
-   the resolved PR base. In a hosted setup, run the resolved-base gate first
-   when the resolved base tracking ref is not `origin/main`. This includes fork
-   and stacked PRs. Then fetch `origin/main` and warm the hook with
-   `./scripts/agent-quality-gate.sh --run --parallel 3 --base origin/main`. If
-   the resolved base is `origin/main`, this hook warm is also the resolved-base
-   gate.
-   Before invoking it, ensure
-   that no direct validation, dashboard server, or browser suite outside the
-   coordinator is active on the same machine. Concurrent `--run` gates from
-   other worktrees can continue through the coordinator. They share weighted
-   machine capacity. From invocation until this gate exits, do not start
-   uncoordinated work there. Use same-machine spare workers only for read-only
-   work. Run validation outside the coordinator from a fully hydrated checkout
-   on another machine.
-4. Report every failed or skipped command, the relevant output, and the
+1. Inspect the branch diff against the resolved pull request base. Use the
+   resolved PR base, not a fixed `origin/main`, for every diff-based author
+   check, including stacked PRs.
+2. Read the scoped instructions and checklists for the changed surfaces.
+3. Select the applicable direct author checks from step 3 of
+   [`pr-operating-card.md`](../../docs/notes/pr-operating-card.md). Unless the
+   requester asked for dry verification only, run those checks directly and
+   record each result as the card requires. Do not substitute the legacy
+   diagnostic gate for the author-check table.
+4. Report every failed or not-run command, the relevant output, and the
    smallest next fix.
 
 ## Output

.agents/skills/doc-garden/SKILL.md

@@ -5,7 +5,7 @@ title: Documentation Garden Skill
 status: active
 owner: eng
 canonical: true
-last_verified: 2026-07-22
+last_verified: 2026-09-02
 doc_type: skill
 scope: repo-wide
 review_interval_days: 90
@@ -89,9 +89,9 @@ pnpm agent:context-check
 pnpm agent:context-budget --strict
 ```
 
-Run the quality gate through step 3 of the operating card. Hosted sessions use
-the exact direct warm command from that step. Local sessions use
-`pnpm agent:quality-gate --run`.
+Apply the direct author checks from step 3 of the
+[operating card](../../../docs/notes/pr-operating-card.md). Record each
+applicable result in the PR validation section as that step requires.
 
 Summarize the disposition and evidence for every packet file in the PR body.
 Open a normal ready-for-review PR, use `Closes #<issue>` only when the entire

.agents/skills/monorepo-import/SKILL.md

@@ -5,7 +5,7 @@ title: Monorepo Import Skill
 status: active
 owner: eng
 canonical: true
-last_verified: 2026-07-23
+last_verified: 2026-09-02
 doc_type: skill
 scope: repo-wide
 review_interval_days: 90
@@ -83,16 +83,17 @@ patterns and update them in the same PR.
 
 ## Phase 4: Verify In Layers
 
-Run the narrow local checks for the imported subsystem first, then the repo's
-agent gate:
+Run the narrow local checks for the imported subsystem first. Then apply the
+repo-wide checks for its changed surfaces:
 
 - package build/typecheck/test/lint for imported TypeScript packages
 - `forge test` for imported Foundry helpers/contracts
 - Terraform fmt/init/validate/plan for touched Terraform roots; never apply
   without explicit user approval
 - workflow syntax and path/filter review for new CI/deploy workflows
-- the quality gate from operating-card step 3 before opening or updating the
-  PR; hosted sessions use its exact direct warm command
+- the applicable direct author checks from step 3 of the
+  [operating card](../../../docs/notes/pr-operating-card.md) before the ready
+  handoff
 
 If registry/network access is unavailable, do not pretend lockfile or install
 work is verified. Surface the blocked command and rerun with network access when

.agents/skills/ship/SKILL.md

@@ -1,11 +1,11 @@
 ---
 name: ship
-description: '[repo-skill] Ship monitoring-monorepo changes through the repo''s Codex-compatible workflow: preflight, quality gate, closeout review, commit, push, PR create/update, readiness babysitting, and required production closeout. Use when the user says "ship it", "/ship", "push this", "open a PR", "create a PR", "publish this", or "send it" in this repo.'
+description: '[repo-skill] Ship monitoring-monorepo changes through the repo''s Codex-compatible workflow: preflight, direct author checks, closeout review, commit, push, PR create/update, readiness babysitting, and required production closeout. Use when the user says "ship it", "/ship", "push this", "open a PR", "create a PR", "publish this", or "send it" in this repo.'
 title: Ship Skill
 status: active
 owner: eng
 canonical: true
-last_verified: 2026-08-22
+last_verified: 2026-09-02
 doc_type: skill
 scope: repo-wide
 review_interval_days: 90
@@ -20,40 +20,40 @@ loads the user-global `ship` skill instead of this file
 ([`codex-agent-skills.md`](../../../docs/notes/codex-agent-skills.md#claude-global-store-shadowing)),
 so a rule written here and nowhere else would reach Codex only.
 
-Work [`pr-operating-card.md`](../../../docs/notes/pr-operating-card.md) from
-step 2 through step 9. It owns the PR description shape, the ready-for-review
-default, `Closes` vs `Refs`, the babysit and ready-state contracts, merge
-hygiene, and production closeout.
+After bootstrap, work
+[`pr-operating-card.md`](../../../docs/notes/pr-operating-card.md) steps 2-9.
+Run its preflight before agent-invoked repository code. It does not attest
+bootstrap; setup has a separate [trust boundary](../../../docs/notes/worktree-and-web-setup.md#bootstrap-trust-boundary).
 
-| Decision                                                             | Authority                                                                                   |
-| -------------------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
-| Gate mapping, autoreview engine, bundle preparation and verification | [`agent-quality-gate-mechanics.md`](../../../docs/notes/agent-quality-gate-mechanics.md)    |
-| Readiness and feedback projections                                   | [`pr-ready-state.md`](../../../docs/notes/pr-ready-state.md)                                |
-| Surface detection, gh capability gate, gh→MCP mapping                | [`github-tooling-surfaces.md`](../../../docs/notes/github-tooling-surfaces.md)              |
-| UI browser verification and the `## Visual comparison` section       | [`dashboard-verification.md`](../../../docs/notes/dashboard-verification.md)                |
-| Claim, deferrals, issue lifecycle                                    | [`agent-issue-workflow.md`](../../../docs/notes/agent-issue-workflow.md)                    |
-| Production closeout                                                  | [`../../../docs/deployment.md`](../../../docs/deployment.md) and the owning package runbook |
+| Decision                                                       | Authority                                                                                   |
+| -------------------------------------------------------------- | ------------------------------------------------------------------------------------------- |
+| Author-check triggers and result records                       | [`pr-operating-card.md`](../../../docs/notes/pr-operating-card.md#the-loop)                 |
+| Autoreview engine, bundle preparation and verification         | [`agent-quality-gate-mechanics.md`](../../../docs/notes/agent-quality-gate-mechanics.md)    |
+| Readiness and feedback projections                             | [`pr-ready-state.md`](../../../docs/notes/pr-ready-state.md)                                |
+| Surface detection, gh capability gate, gh→MCP mapping          | [`github-tooling-surfaces.md`](../../../docs/notes/github-tooling-surfaces.md)              |
+| UI browser verification and the `## Visual comparison` section | [`dashboard-verification.md`](../../../docs/notes/dashboard-verification.md)                |
+| Claim, deferrals, issue lifecycle                              | [`agent-issue-workflow.md`](../../../docs/notes/agent-issue-workflow.md)                    |
+| Production closeout                                            | [`../../../docs/deployment.md`](../../../docs/deployment.md) and the owning package runbook |
 
 ## What this repo adds
 
-- **Gate before publish**: Follow card step 3 and validate against the resolved
-  PR base. A local setup runs `pnpm agent:quality-gate --run` against that base.
-  In a hosted setup, run the resolved-base gate first when the resolved base
-  tracking ref is not `origin/main`. This includes fork and stacked PRs. Then
-  fetch `origin/main` and warm the hook with
-  `./scripts/agent-quality-gate.sh --run --parallel 3 --base origin/main`. If the
-  resolved base is `origin/main`, this hook warm is also the resolved-base
-  gate. Then run the closeout review
-  for non-trivial behavioural, workflow, security, data-flow, infrastructure,
-  or UI changes. **Which closeout depends on the surface**, and a bare
+- **Repository identity before agent-invoked code**: Resolve `CURRENT_REPO`,
+  `BASE_REPO`, PR, remotes, and base after bootstrap. Stop fork and
+  cross-repository heads before agent code. Bind PR probes to `BASE_REPO`; a
+  failed query does not prove that no PR exists. Use the resolved PR base, not
+  a fixed `origin/main`, for every diff-based author check, including stacked
+  PRs.
+- **Author checks before ready publication**: Apply every matching row in card
+  step 3. Format every intended changed file before those checks. Record each
+  result in `## Validation`. Reapply affected rows after a material fix, base
+  integration, or formatter change during commit. Pre-push runs no repository
+  verification.
+  Then run the closeout review for non-trivial behavioural, workflow, security,
+  data-flow, infrastructure, or UI changes. **Which closeout depends on the surface**, and a bare
   `pnpm agent:autoreview` is not always it: inside an active Codex session that
   silently selects the local deterministic engine, so the prepared-bundle
   fresh-context flow is required instead. Card step 4 owns the choice — follow
   it rather than the bare command.
-- **Resolve the base repo from evidence.** A fork checkout uses its parent as
-  `BASE_REPO`; never substitute a fork's `origin` for its parent. Bind every
-  `gh pr view`, feedback-state, and ready-state call with `--repo <BASE_REPO>`.
-  A failed GitHub query is not evidence that no PR exists.
 - **PRs open ready for review.** Drafts suppress the automated AI reviews this
   workflow depends on.
 - **`scripts/pr/check-pr-description.mjs` enforces `## The Problem` then
@@ -63,8 +63,8 @@ hygiene, and production closeout.
 - **Never post routine or duplicate `@codex review` requests**, and never tag
   `chatgpt-codex-connector` directly.
 - **Deep security scan** (`claude-security`) is developer-installed and Claude
-  Code only. This repo does not declare it. Where unavailable, aim the gate and
-  closeout review at the sensitive surfaces and record
+  Code only. This repo does not declare it. Where unavailable, aim the direct
+  author checks and closeout review at the sensitive surfaces and record
   `Claude Security scan: skipped (<surface>)` in the final summary.
 - **Done is not merge when Done means includes live behaviour.** Card step 9
   owns the closeout.

.claude/commands/autoreview.md

@@ -24,7 +24,8 @@ checks and autoreview for that batch. Do not pause solely for cycle count before
 five review-triggered patch cycles are complete; pause for scope
 reclassification before starting a sixth. A clean source review is not test,
 browser, generated-artifact, CLI/API, or runtime proof, so retain every
-applicable gate.
+applicable author check from step 3 of the
+[operating card](../../docs/notes/pr-operating-card.md).
 If an autoreview runtime change triggers the owning adapter's self-review
 refusal, keep it intact and follow the trusted pre-change sequence in the owner
 note.

.claude/hooks/session-start.sh

@@ -28,13 +28,6 @@ fi
 REPO_ROOT="${CLAUDE_PROJECT_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)}"
 SETUP_SCRIPT="$REPO_ROOT/scripts/bootstrap/claude-code-web-setup.sh"
 
-# Cached remote sessions can resume after this hook updates but before the
-# heavy startup setup runs again. Apply the lightweight hosted pre-push policy
-# on every remote source before the resume/compact early exit.
-if git -C "$REPO_ROOT" rev-parse --git-dir >/dev/null 2>&1; then
-  git -C "$REPO_ROOT" config agent.qualityGate.cloudPrePushRequireFresh true
-fi
-
 # Inspect SessionStart payload on stdin (best-effort: if stdin is not present
 # or the payload is unparsable, default to running the bootstrap rather than
 # silently skipping). Only `startup` triggers the heavy install/codegen path.