AI-ops audit P0s: re-tier agents, live gates, one rule source

.claude/agents/build-monitor.md

@@ -12,4 +12,4 @@ Use skills (minimal subset):
 - `best-practices`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/content-creator.md

@@ -1,7 +1,7 @@
 ---
 name: content-creator
 description: Creates or edits site/course content under the JT voice guide, canon, and anti-AI-pattern gates. Voice-sensitive judgment work.
-model: fable
+model: opus
 ---
 
 # Content Creator (Canonical)
@@ -15,6 +15,6 @@ Use skills (minimal subset):
 - `copywriting-core`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 - For full blog workflow, see `docs/workflows/blog-pipeline.md`
 - For cover images, follow `docs/workflows/cover-images.md` and `.stitch/design.md`

.claude/agents/core/analyst.md

@@ -13,4 +13,4 @@ Use skills (minimal subset):
 - `six-thinking-hats`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/core/coder.md

@@ -1,7 +1,7 @@
 ---
 name: core-coder
-description: Implements changes with minimal context - thin wrapper over the incremental-implementation skill set and BASE_HANDBOOK rules.
-model: fable
+description: Implements changes with minimal context - thin wrapper over the incremental-implementation skill set and CLAUDE.md rules.
+model: opus
 ---
 
 # Core Coder (Critical)
@@ -13,7 +13,7 @@ Use skills (minimal subset):
 - `best-practices`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 - If a task is domain-specific (SEO/Hugo/content), defer to the canonical agent or skill
 
 ## Methodology stance (XP / lean)

.claude/agents/core/planner.md

@@ -13,7 +13,7 @@ Use skills (minimal subset):
 - `incremental-implementation`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 
 ## Methodology stance (Shape Up / lean)
 

.claude/agents/core/researcher.md

@@ -13,7 +13,7 @@ Use skills (minimal subset):
 - `knowledge-synthesis`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 - Obey the Research Protocol for claude-context searches
 
 ## Methodology stance (lean)

.claude/agents/core/reviewer.md

@@ -1,7 +1,7 @@
 ---
 name: core-reviewer
 description: Reviews changes for correctness and risk across the repo - the general 4-eyes gate. Judgment role, findings with file:line evidence.
-model: fable
+model: opus
 ---
 
 # Core Reviewer (Critical)
@@ -13,7 +13,7 @@ Use skills (minimal subset):
 - `code-review-quality`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 
 ## Methodology stance (XP / lean)
 

.claude/agents/core/screenshot-guardian.md

@@ -13,4 +13,4 @@ Use skills (minimal subset):
 - `pagespeed-insights`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/core/tester.md

@@ -14,7 +14,7 @@ Use skills (minimal subset):
 - `tdd-fix-tests`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 
 ## Methodology stance (XP)
 

.claude/agents/hugo-expert.md

@@ -14,5 +14,5 @@ Use skills (minimal subset):
 - `postcss-best-practices`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 - For CSS rules see `docs/workflows/css-consolidation.md`

.claude/agents/okf-maintainer.md

@@ -0,0 +1,42 @@
+---
+name: okf-maintainer
+description: Maintains the .okf/ knowledge bundle - distills a durable learning into its concept file + section index + a dated log.md entry, riding the SAME commit/branch as the change that produced it. Dispatch in parallel with the main work per the ENFORCED CLAUDE.md OKF rule. Also upserts the concept's okf-* mirror entry in ruflo memory (namespace okf) per the sync rule.
+model: opus
+---
+
+# OKF Maintainer
+
+You maintain `.okf/` — the Open Knowledge Format bundle. You are dispatched
+with (a) the durable learning (root cause, gotcha, decision, canon change) and
+(b) the branch/commit it must ride with.
+
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
+- Consume the bundle via the `okf:okf` skill; never hand-edit the bundle FORMAT
+  from memory — the skill carries the spec.
+
+## The job, every time
+
+1. **Consolidate before creating**: find the existing concept the learning
+   belongs to (`.okf/*/index.md` first). A new concept file is the exception
+   and needs a matching index row.
+2. Update the concept + its section `index.md` row if the description changed.
+3. Append a dated `## YYYY-MM-DD - <headline>` entry to `.okf/log.md`
+   (newest first) — what changed, why, and the pointer.
+4. **Ruflo mirror (sync rule)**: if a concept's title/description changed,
+   upsert its memory entry:
+   `npx @claude-flow/cli@latest memory store -n okf -k okf-<section-abbrev>-<slug> -v "<Type>: <description> Details: <path>"`
+   (prefixes: okf-build- / okf-content- / okf-wf- / okf-design- / okf-arch- / okf-cs-).
+5. Validate: run the `okf:validate` skill's checker with `--strict`; the gate
+   is **0 ERRORS** (warnings are a known-red state — check the error count,
+   not the exit code or the checkmark).
+6. Commit on the SAME branch as the main work (or hand the staged files back
+   to the coordinator if it owns the commit) — the bundle update rides the
+   same commit as the change. Never a separate cleanup PR.
+
+## Boundaries
+
+- You edit `.okf/` and the ruflo okf namespace only. Never the change under
+  documentation, never CLAUDE.md, never STATUS.md — flag mismatches back to
+  the coordinator instead.
+- One learning per dispatch; if you find a second stale thing while working,
+  report it, don't fix it silently.

.claude/agents/ruby/ruby-coder.md

@@ -1,7 +1,7 @@
 ---
 name: ruby-coder
 description: Ruby/test-suite changes with minimal context - Minitest, Rake tasks, bin/ scripts. Follows 37signals/DHH style and the repo's TDD gates.
-model: sonnet
+model: opus
 ---
 
 # Ruby Coder (Critical)
@@ -14,4 +14,4 @@ Use skills (minimal subset):
 - `minitest`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/ruby/ruby-reviewer.md

@@ -1,7 +1,7 @@
 ---
 name: ruby-reviewer
 description: Reviews Ruby/test-suite changes for correctness, risk, and repo-convention fit. The 4-eyes gate for Ruby work - judgment role, evidence-cited findings.
-model: fable
+model: opus
 ---
 
 # Ruby Reviewer (Critical)
@@ -13,4 +13,4 @@ Use skills (minimal subset):
 - `rails-expert`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/seo-specialist.md

@@ -15,5 +15,5 @@ Use skills (minimal subset):
 - `pagespeed-insights`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`
 - If the task is not SEO-related, defer to the appropriate skill or agent

.claude/agents/validation/qa-browser-tester.md

@@ -13,4 +13,4 @@ Use skills (minimal subset):
 - `web-design-reviewer`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/validation/ui-problem-diagnosis-specialist.md

@@ -13,4 +13,4 @@ Use skills (minimal subset):
 - `ux-principles`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/agents/validation/ux-browser-validator.md

@@ -13,4 +13,4 @@ Use skills (minimal subset):
 - `ux-principles`
 
 Rules:
-- Follow `docs/workflows/BASE_HANDBOOK.md`
+- Follow `CLAUDE.md` (repo root — the canonical rules) and `docs/workflows/flow-router.md`

.claude/settings.json

@@ -8,6 +8,11 @@
             "type": "command",
             "command": "bin/agent-bootstrap > /tmp/agent-bootstrap.log 2>&1 || true",
             "timeout": 600
+          },
+          {
+            "type": "command",
+            "command": "npx @claude-flow/cli@latest memory search -q CORRECTION --type keyword --limit 5 2>/dev/null || true",
+            "timeout": 45
           }
         ]
       }

.github/workflows/publish.yml

@@ -89,6 +89,14 @@ jobs:
           PRECOMPILED_ASSETS: '1'
           HUGO_DEFAULT_PATH: _dest/public-test
 
+      # Course content validators (chapter numbering, links, em-dash,
+      # year-stamp...). Previously only wired to an uninstalled pre-push hook —
+      # a green config test stood in front of a gate that never ran
+      # (docs/70-79-ai-intelligence/70.12, P0). CI is where it cannot be
+      # uninstalled.
+      - name: Course validators
+        run: bin/validate-course
+
   # Asset-pipeline integration: dev vs prod branching in css-processor,
   # css-inline, and js-processor (fingerprint/integrity/minify/crossorigin).
   # Previously ran only inside `rake test` on push-to-master, so a PR could

.github/workflows/test.yml

@@ -9,8 +9,8 @@ name: Screenshot Tests
 # PR #386) - and record mode would silently overwrite the canonical
 # baselines with unpinned renders.
 #
-# PR runs are REPORT-ONLY during the soak period (continue-on-error) -
-# flip that flag off to make the gate blocking.
+# PR runs are BLOCKING (2026-08-22): the report-only soak that started
+# 2026-07-31 ended after its motivating defects were fixed.
 
 on:
   workflow_dispatch:
@@ -67,9 +67,10 @@ jobs:
     if: ${{ github.event_name == 'pull_request' || inputs.screenshots || inputs.update-baselines }}
     runs-on: ubuntu-latest
     timeout-minutes: 20
-    # Report-only soak for the PR gate: failures upload the diff report and
-    # comment on the PR but do not block merging. Remove after the soak.
-    continue-on-error: ${{ github.event_name == 'pull_request' }}
+    # BLOCKING since 2026-08-22: the report-only soak (2026-07-31, #413) ended —
+    # its motivating defects were fixed (#560 tolerance, #566 baselines, #570
+    # per-page pins) and 22 days report-only meant nothing blocking checked
+    # Linux pixels (dtest-from-worktree is vacuous).
 
     steps:
       # PR events use the default checkout (the PR *merge* commit): testing

.okf/log.md

@@ -111,6 +111,24 @@ Also corrected: the ~1e-6 noise figure in test-gates and in the code comment
 was macOS-local and read as universal. Same overclaim shape the #566 reviewer
 caught one layer down.
 
+## 2026-08-22 - audit P0s executed: re-tiered agents, live gates, one rule source
+
+All 70.12 P0s landed the same day. Five agents re-tiered off fable/sonnet to
+opus (core coder+reviewer, ruby pair, content-creator) - the reviewer
+quota-death root cause. bin/validate-course now runs in CI's unit job and the
+config test that green-covered its uninstalled pre-push hook is DELETED (the
+enforcement lens's own dissent: fix by deletion, not elaboration). The CI
+visual gate's 22-day report-only soak ended - continue-on-error removed.
+BASE_HANDBOOK.md deleted with a full reference sweep; CLAUDE.md is the one
+rule source and all 20 agents now cite it. okf-maintainer exists as a real
+agent (the ENFORCED rule's phantom executor, built); the phantom "weekly
+maintain pass" claim removed from both rule copies. Docs-CI three-way
+contradiction resolved (no CI fires on docs-only; none is waited for). mem0
+dropped (plugin off, fallback = memoria); machine config deduped (~100k
+schema tokens/session: ruflo-core plugin MCP, chrome-devtools x2, github,
+serena, context7 x2, notebooklm). Ruflo retrieval wired into SessionStart
+for the 30-day read measurement. Full audit: docs/70-79-ai-intelligence/70.12.
+
 ## 2026-08-22 - six-lens AI-ops audit: the failure mode is silent no-ops, not missing rules
 
 A /swarm panel (six read-only lenses, mandatory dissents) audited the AI

.okf/workflows/company-layer-ownership.md

@@ -58,7 +58,7 @@ the OKF same-commit wording is the pattern that has held).
 
 # Where the routers point
 
-Aligned 2026-08-17: `AGENTS.md`, `docs/workflows/BASE_HANDBOOK.md`,
+Aligned 2026-08-17: `AGENTS.md`,
 `docs/workflows/flow-router.md` (split into two routes), `CLAUDE.md`,
 `docs/projects/2607-vibe-code-rescue/strategy.md`, and
 `.agents/skills/async-first-communication/SKILL.md`. If a new doc tells an

.okf/workflows/ruflo-runtime.md

@@ -55,6 +55,19 @@ per concept from its frontmatter title/description, exactly as above.
   by design has no pattern tables - the skip message "agentdb schema not
   initialised" means you pointed it at the wrong DB, not that init is needed.
 - 12 background workers exist (`hooks worker list`); the daemon runs them.
+- **Retrieval is wired (2026-08-22)**: a second SessionStart hook runs
+  `memory search -q CORRECTION --type keyword --limit 5` unredirected, so
+  recalled corrections inject into session context AND increment
+  `access_count` — the 30-day read measurement that decides the memory
+  architecture (audit doc 70.12). Do not remove it before the measurement.
+- **ruflo-core PLUGIN is disabled (2026-08-22)** — it bundled a duplicate MCP
+  registration (~61.5k schema tokens/session); the project-level `ruflo`
+  entry in `~/.claude.json` is the ONE registration (CLAUDE.md hardcodes its
+  `mcp__ruflo__*` prefix). Its skills (ruflo-doctor, init-project, witness)
+  were thin wrappers over `npx @claude-flow/cli@latest ...` — call the CLI
+  directly (doctor / init upgrade / etc.). Other ruflo-* plugins stay enabled.
+- **mem0 is dropped from the stack (Paul 2026-08-22)** — plugin disabled;
+  quota-dead and redundant. Fallback chain is ruflo → memoria.
 
 # CLI bugs observed 2026-08-22 (ruflo v3.38.16) - do not re-fight these
 
@@ -64,10 +77,13 @@ or still present.
 
 - `config init` generates a config the loader rejects ("reading 'map'"
   warning on every call). Run on defaults; delete any generated config.
-- `memory export` fails ("Exported to undefined"); `memory_import_claude`
-  finds 0 files even with `allProjects: true` (path-encoding). The auto-memory
-  corpus is deliberately NOT mirrored into ruflo - `MEMORY.md` already loads
-  every session.
+- `memory export` fails ("Exported to undefined"). `memory_import_claude`
+  has a path-encoding bug (replaces only slashes, but Claude's memory dir
+  name also dashes the dots) — WORKAROUND: pass a pre-dashed projectPath,
+  e.g. `/Users/pftg/dev/jetthoughts-github-io`, and it imports. The
+  auto-memory corpus (48 files) IS mirrored via that workaround since
+  2026-08-22, namespace `claude-memories`, semantic recall verified —
+  re-import after big memory-file batches (idempotent, dedups).
 - CLI `memory search --type semantic` returns empty in terminals (needs
   @ruvector/core); single-term `--type keyword` works; real semantic search is
   MCP-only. (Also in CLAUDE.md research protocol.)

AGENTS.md

@@ -1,15 +1,15 @@
 # AGENTS.md — JetThoughts Blog
 
 Canonical instructions live in `CLAUDE.md`. Read that first.
-Session start: always read `@docs/workflows/BASE_HANDBOOK.md` and `@docs/workflows/flow-router.md`.
+Session start: always read `@docs/workflows/flow-router.md` (rules live in `CLAUDE.md`).
 
 **Async-first communication (SOP for ALL agents):** written, discoverable artifacts are the default for every decision, finding, status change, and handoff — sync interaction is the exception and its outcome gets written back same-day. Full rules + canonical surfaces table: `.agents/skills/async-first-communication/SKILL.md`. A task is not done until its state is readable asynchronously. **Cold-start surface: `STATUS.md` at repo root** (goals + WIP + blockers, links only) — a session that changed what's in flight updates it in the same commit/PR, or the session is not done.
 
-**Four eyes on every stage, author != verifier (SOP for ALL agents):** the point is agents collaborating to find the TRUTH — a single agent cannot find its own blind spot, because it checks the thing it MEANT to build. Every change is **written by one sub-agent and verified by a different one**; the author never produces the evidence for its own claim. The review lands before the artifact leaves the workshop: a plan is peer-reviewed BEFORE the user is asked, a diff BEFORE commit, a finding is reproduced before it is reported, a measurement re-derived before it is quoted. Brief reviewers with the goal and the artifact, never your conclusions; ask for measurements, not verdicts. Routing: internal sub-agents for every per-stage review, `/codex:review` ONCE at the final verify before merge for USER-FACING changes (it is slow — never in the inner loop); docs-only and instruction-layer changes ship on internal review + CI and apply its findings as a follow-up. No reviewer tooling in this runtime? Fall back agent → external → peer session → human, and mark the change UNREVIEWED if none is reachable. Full protocol: the `jt-delivery:contract` skill (plugin `jt-delivery@jetthoughts`); repo bindings: `docs/workflows/autonomous-delivery-prompt.md`.
+**Four eyes on every stage, author != verifier (SOP for ALL agents):** the point is agents collaborating to find the TRUTH — a single agent cannot find its own blind spot, because it checks the thing it MEANT to build. Every change is **written by one sub-agent and verified by a different one**; the author never produces the evidence for its own claim. The review lands before the artifact leaves the workshop: a plan is peer-reviewed BEFORE the user is asked, a diff BEFORE commit, a finding is reproduced before it is reported, a measurement re-derived before it is quoted. Brief reviewers with the goal and the artifact, never your conclusions; ask for measurements, not verdicts. Routing: internal sub-agents for every per-stage review, `/codex:review` ONCE at the final verify before merge for USER-FACING changes (it is slow — never in the inner loop); docs-only and instruction-layer changes ship on internal review + local gates — no CI fires on them (path filters) and none is waited for (Paul 2026-08-22); findings apply as a follow-up. No reviewer tooling in this runtime? Fall back agent → external → peer session → human, and mark the change UNREVIEWED if none is reachable. Full protocol: the `jt-delivery:contract` skill (plugin `jt-delivery@jetthoughts`); repo bindings: `docs/workflows/autonomous-delivery-prompt.md`.
 
 **Session memory (ruflo-first, SOP for ALL agents):** search ruflo memory (`mcp__ruflo__memory_search`, MCP path) before non-obvious decisions; store decisions/corrections via `mcp__ruflo__memory_store` as they land. Details + prefix conventions: `CLAUDE.md` §Memory.
 
-**OKF maintenance (ENFORCED, parallel by default):** durable learnings (canon numbers, root causes, workflow gotchas) go into `.okf/` — dispatch an okf-maintainer in parallel with the main work, not after it; the bundle update rides the same commit as the change. Weekly scheduled maintain pass back-fills misses. See CLAUDE.md OKF section for the full rule.
+**OKF maintenance (ENFORCED, parallel by default):** durable learnings (canon numbers, root causes, workflow gotchas) go into `.okf/` — dispatch the `okf-maintainer` agent in parallel with the main work, not after it; the bundle update rides the same commit as the change. No scheduled back-fill exists; the same-commit rule is the only mechanism. See CLAUDE.md OKF section for the full rule.
 
 <skills>
 - `.agents/skills/async-first-communication/SKILL.md` — async-first knowledge sharing SOP (default for ALL work)