diff --git a/.github/trigger-deploy b/.github/trigger-deploy
index 5aef4763..3a71c762 100644
--- a/.github/trigger-deploy
+++ b/.github/trigger-deploy
@@ -3,3 +3,4 @@
 2026-07-17T17:20:00Z deploy manifest-merge + alias collapse
 2026-07-17T18:00:00Z deploy price tiebreak
 2026-07-18T02:00:00Z deploy wish-bridge qualified-demand gate
+2026-07-18T02:45:00Z deploy operator demand board
diff --git a/.github/workflows/secret-scan.yml b/.github/workflows/secret-scan.yml
new file mode 100644
index 00000000..70871b52
--- /dev/null
+++ b/.github/workflows/secret-scan.yml
@@ -0,0 +1,38 @@
+# Always-on secret scanning. Unlike deploy.yml (marker-gated), this runs on
+# EVERY push to a tracked branch and EVERY pull request — a committed
+# credential should never depend on remembering a marker. gitleaks scans the
+# working tree against its default high-signal ruleset (OpenAI sk- keys,
+# GitHub/AWS tokens, private keys, mnemonics, ...) plus the repo allowlist in
+# .gitleaks.toml, which suppresses this repo's structural false positives
+# (public on-chain addresses, the demo JWT, test canaries). Only a genuinely
+# new secret fails the job.
+name: secret-scan
+
+on:
+  push:
+    branches: ["claude/sweet-brown-i99jl3", "main"]
+  pull_request:
+
+permissions:
+  contents: read
+
+jobs:
+  gitleaks:
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    steps:
+      - uses: actions/checkout@93cb6efe18208431cddfb8368fd83d5badbf9bfd
+      - name: Scan for committed secrets (gitleaks, pinned + checksum-verified)
+        env:
+          GITLEAKS_VERSION: "8.21.2"
+        run: |
+          set -euo pipefail
+          FN="gitleaks_${GITLEAKS_VERSION}_linux_x64.tar.gz"
+          BASE="https://github.com/gitleaks/gitleaks/releases/download/v${GITLEAKS_VERSION}"
+          curl -sSL --fail -o "/tmp/${FN}" "${BASE}/${FN}"
+          curl -sSL --fail -o /tmp/gitleaks_checksums.txt "${BASE}/gitleaks_${GITLEAKS_VERSION}_checksums.txt"
+          ( cd /tmp && grep " ${FN}\$" gitleaks_checksums.txt | sha256sum -c - )
+          tar -xzf "/tmp/${FN}" -C /tmp gitleaks
+          # Tree scan (--no-git): fails the job on any finding not allowlisted
+          # in .gitleaks.toml. --redact keeps any match out of the public logs.
+          /tmp/gitleaks detect --no-git --config .gitleaks.toml --redact --verbose
diff --git a/.gitleaks.toml b/.gitleaks.toml
new file mode 100644
index 00000000..9109dc96
--- /dev/null
+++ b/.gitleaks.toml
@@ -0,0 +1,39 @@
+# gitleaks config for Agent402. Extends the default ruleset (which catches
+# real high-signal secrets: OpenAI sk- keys, GitHub tokens, AWS keys, private
+# keys, mnemonics, etc.) and allowlists the structural false positives this
+# repo carries by design, so CI stays green until a GENUINELY new secret lands.
+#
+# Every allowlist entry below was verified against a full-tree + full-history
+# scan (2026-07-18): all pre-existing hits are public on-chain addresses, the
+# standard demo JWT, or fake test/canary fixtures. Real secrets live only in
+# Railway env and GitHub Actions secrets, never in the repo.
+[extend]
+useDefault = true
+
+[allowlist]
+description = "Agent402 known false positives: on-chain addresses, demo JWTs, test fixtures"
+# Fixtures and copy that intentionally contain fake or public credential-shaped
+# strings. Test files assert on canaries by design; docs carry example tokens.
+paths = [
+  '''scripts/test-.*\.js''',
+  '''docs/.*''',
+  '''package-lock\.json''',
+  '''\.gitleaks\.toml''',
+]
+# Matched against each finding's captured secret.
+regexes = [
+  # EVM contract/wallet addresses are public identifiers, never secrets. This
+  # is 40 hex chars precisely; a real EVM private key is 64 hex and is NOT
+  # covered here, so it would still fire.
+  '''0x[0-9a-fA-F]{40}\b''',
+  # The leak-guard test's own canary tokens (e.g. fredv2-LEAKCANARY0000).
+  '''LEAKCANARY''',
+  # Placeholder KDF-output example in the crypto-hash tool docs.
+  '''a1b2c3d4e5f6''',
+]
+# The standard demo JWT ({"alg":"HS256","typ":"JWT"} header) used as the
+# decode-jwt tool's worked example in kit.js / kit2.js / skills.js. Stopwords
+# match when the secret contains the substring.
+stopwords = [
+  '''eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9''',
+]