merge: sync upstream v0.41.0

Merge kenn-io/agentsview main (v0.41.0, 138 commits ahead of our
0.38.1 fork point) and realign the fork onto the upstream superset.

What this fork gains from upstream:
- Nine new agent parsers, the Recall subsystem, the artifact machine,
  and the Money/microdollars pricing rework across SQLite, PostgreSQL,
  and DuckDB stores.
- Token usage view reworked as the in-page Cost|Tokens toggle with
  four-way token-type filtering, URL state, and an automatic redirect
  from the legacy /token-usage route (upstream kenn-io#1234, authored by this
  fork's maintainer).
- Kimi Work fixes beyond the fork copy: tool-step usage back-fill,
  k2d6-agent K2.6 pricing, cwd extraction, and date-ambiguous alias
  pricing behind the Kimi model-era cutoff (kenn-io#1338, kenn-io#1427).

Conflict resolution across 40 files:
- Token usage UI, Kimi Work parser/pricing, usage backends, docs, and
  golden fixtures: take upstream. The fork's older components predate
  the Money response protocol and cannot coexist with it.
- start-agentsview.command and start-agentsview-dev.command: keep the
  fork versions, which carry later local refinements (Makefile
  alignment, link to the local build doc) missing upstream.
- 本地构建说明.md is the only other fork-only file; it merged without
  conflict. Golden v2 fixtures follow upstream's v4/v5 contract rename.

First run after this merge reparses stored sessions (data version 70
to 88, non-destructive) and rebuilds any DuckDB mirror under the new
schema version by design.

Author: zhangyu

.github/actions/build-desktop-artifact/action.yml

@@ -14,18 +14,20 @@ inputs:
   artifact_dir:
     description: Bundle output directory
     required: true
-
 runs:
   using: composite
   steps:
-    - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16  # v6.5.0
+    - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e  # v7.0.0
       with:
         go-version-file: go.mod
+        cache: ${{ inputs.bundle == 'appimage' && inputs.target_triple == '' && github.repository == 'kenn-io/agentsview' && (github.event_name == 'push' && github.ref == 'refs/heads/main' || (github.event_name == 'pull_request' && github.event.pull_request.head.repo.full_name == github.repository && github.event.pull_request.base.repo.full_name == github.repository)) && 'false' || 'true' }}
 
-    - uses: actions/setup-node@48b55a011bda9f5d6aeb4c2d9c7362e8dae4041e  # v6.4.0
+    - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020  # v7.0.0
       with:
         node-version: "24"
 
+    - uses: dtolnay/rust-toolchain@4360b52568e2003a75bf9bc1d59f33a8e3fc893c  # stable
+
     - name: Setup MinGW (Windows)
       if: runner.os == 'Windows'
       uses: msys2/setup-msys2@66cd2cce69caa17b53920067426061ca1de3a884  # v2
@@ -73,7 +75,7 @@ runs:
         cd desktop
         export TAURI_ENV_TARGET_TRIPLE="$AGENTSVIEW_TARGET_TRIPLE"
         npm run prepare-sidecar
-        build_cmd=(npx tauri build --bundles "$BUNDLE")
+        build_cmd=(npx tauri build --verbose --bundles "$BUNDLE")
         if [ -n "$AGENTSVIEW_TARGET_TRIPLE" ]; then
           build_cmd+=(--target "$AGENTSVIEW_TARGET_TRIPLE")
         fi

.github/workflows/bench-pr.yml

@@ -0,0 +1,21 @@
+name: Bench Gate
+
+# The PR dispatcher always calls the main-pinned workflow, which runs
+# on GitHub-hosted runners for every PR.
+on:
+  pull_request:
+    # Docs/frontend-only PRs cannot change the gated Go paths. If
+    # this check is ever made required on branch protection, pair it
+    # with a no-op sibling workflow on the inverse paths.
+    paths:
+      - "**.go"
+      - "go.mod"
+      - "go.sum"
+      - "Makefile"
+      - ".github/workflows/bench.yml"
+      - ".github/workflows/bench-pr.yml"
+permissions: read-all
+
+jobs:
+  run:
+    uses: kenn-io/agentsview/.github/workflows/bench.yml@main

.github/workflows/bench.yml

@@ -1,19 +1,20 @@
 name: Bench Gate
 
 # Compares hot-path benchmarks between the PR head and its merge
-# base, so performance regressions in the sync engine and DB write
-# paths fail the PR instead of shipping. The gated regression classes
-# have all shipped before:
+# base, so performance regressions in sync, DB access, secret scanning,
+# and signal analysis fail the PR instead of shipping. The gated
+# regression classes have all shipped before:
 #
 #   - discovery/skip work scaling with archive size instead of new
 #     data (#912, the providerSourceUnchangedInDB gap)
 #   - O(session history) work per incremental append (#954)
 #   - bulk ingest throughput (#411)
 #   - per-row query-shape regressions in usage aggregation (#309)
+#   - repeated token-set construction during duplicate-prompt analysis
 #
 # Both sides run `make bench-gate` — the Makefile is the single
 # source of truth for the gated package list, sample count, and
-# iteration count — and cmd/benchgate compares the outputs. It gates
+# per-tier iteration counts — and cmd/benchgate compares the outputs. It gates
 # on allocs/op and B/op (deterministic on a given machine, tight
 # thresholds) and on ns/op with a loose 2x threshold that only
 # catches algorithmic blowups; both sides run on the same runner
@@ -28,17 +29,7 @@ name: Bench Gate
 # produced still gate) rather than silently disabling the whole gate.
 
 on:
-  pull_request:
-    # Docs/frontend-only PRs cannot change the gated Go paths. If
-    # this check is ever made required on branch protection, pair it
-    # with a no-op sibling workflow on the inverse paths.
-    paths:
-      - "**.go"
-      - "go.mod"
-      - "go.sum"
-      - "Makefile"
-      - ".github/workflows/bench.yml"
-
+  workflow_call:
 concurrency:
   group: bench-${{ github.head_ref || github.ref }}
   cancel-in-progress: true
@@ -49,14 +40,18 @@ permissions:
 jobs:
   bench-gate:
     name: Benchmark Gate
+    # Always GitHub-hosted, unlike the rest of CI: both sides of the
+    # comparison run on the same runner within one job, so absolute
+    # speed is what matters, and ubuntu-latest runs this I/O-heavy
+    # workload 2-4x faster than the managed fleet.
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7.0.0
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1  # v7.0.1
         with:
           persist-credentials: false
           fetch-depth: 0
 
-      - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16  # v6.5.0
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e  # v7.0.0
         with:
           go-version-file: go.mod
 
@@ -75,11 +70,13 @@ jobs:
         # exists on both sides. The warning makes the degraded run
         # visible instead of a silently green vacuous pass.
         #
-        # The sample count and fixed iteration count are evaluated
-        # from the PR head's Makefile and passed into the base run:
-        # two benchmarks grow their fixture per iteration, so a PR
-        # that changes BENCH_GATE_COUNT/TIME must not compare against
-        # a baseline measured with the old values. The package list
+        # The sample count, fixed iteration counts, and heavy-tier
+        # regex are evaluated from the PR head's Makefile and passed
+        # into the base run: two benchmarks grow their fixture per
+        # iteration, so a PR that changes BENCH_GATE_COUNT/TIME or
+        # moves a benchmark between tiers must not compare against a
+        # baseline measured with the old values. A base Makefile that
+        # predates a variable simply ignores it. The package list
         # intentionally stays per-side, so growing the gate cannot
         # break the base run.
         run: |
@@ -89,7 +86,9 @@ jobs:
           git worktree add /tmp/bench-base "$base"
           if ! make -s -C /tmp/bench-base bench-gate \
               BENCH_GATE_COUNT="$BENCH_GATE_COUNT" \
-              BENCH_GATE_TIME="$BENCH_GATE_TIME" > /tmp/bench-old.txt; then
+              BENCH_GATE_TIME="$BENCH_GATE_TIME" \
+              BENCH_GATE_HEAVY="$BENCH_GATE_HEAVY" \
+              BENCH_GATE_HEAVY_TIME="$BENCH_GATE_HEAVY_TIME" > /tmp/bench-old.txt; then
             echo "::warning title=Bench Gate::merge-base benchmark run exited non-zero; gating against its partial output"
           fi
 

.github/workflows/ci-macos-main.yml

@@ -6,6 +6,13 @@ on:
   # this exact workflow path only from refs/heads/main.
   push:
     branches: [main]
+    paths:
+      - 'cmd/**'
+      - 'internal/**'
+      - 'testdata/golden/**'
+      - 'go.mod'
+      - 'go.sum'
+      - '.github/workflows/ci-macos-main.yml'
 
 permissions:
   contents: read
@@ -21,12 +28,12 @@ jobs:
       group: ci-runners
       labels: [self-hosted, macOS, ARM64]
     steps:
-      - uses: actions/checkout@9c091bb21b7c1c1d1991bb908d89e4e9dddfe3e0  # v7.0.0
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1  # v7.0.1
         with:
           fetch-depth: 0
           persist-credentials: false
 
-      - uses: actions/setup-go@924ae3a1cded613372ab5595356fb5720e22ba16  # v6.5.0
+      - uses: actions/setup-go@b7ad1dad31e06c5925ef5d2fc7ad053ef454303e  # v7.0.0
         with:
           go-version-file: go.mod
 

.github/workflows/ci-pr.yml

@@ -0,0 +1,13 @@
+name: CI
+
+# The PR dispatcher always calls the main-pinned workflow. That workflow
+# independently routes same-repository Linux jobs to the managed public fleet
+# and fork jobs to GitHub-hosted runners.
+on:
+  pull_request:
+
+permissions: read-all
+
+jobs:
+  run:
+    uses: kenn-io/agentsview/.github/workflows/ci.yml@main