feat/work-modes — phase 002 closed, phase 003 opened, and the store the code actually reads

.github/workflows/ci.yml

@@ -1,5 +1,10 @@
 name: tests
 
+# This job already tests the MERGE RESULT: `actions/checkout` resolves
+# `refs/pull/N/merge` on a `pull_request` event, not the branch head. What it
+# cannot see is a base that moves after the run, or a second PR open at the
+# same time — that is `merge-check.yml`, which explains both in full.
+
 on:
   push:
     branches: [main]
@@ -13,7 +18,8 @@ jobs:
       - uses: actions/setup-python@v5
         with:
           python-version: "3.11"
-      # No dependencies to install — Perry's scripts and tests are stdlib only.
-      # The viewer's Flask requirement is opt-in and not exercised here.
+      # No dependencies to install — Perry's scripts and tests are stdlib only,
+      # with no exception since TASK-178 deleted the web viewer (its Flask,
+      # markdown and jinja2 were the only non-stdlib packages Perry ever named).
       - name: Run the suite
         run: bash tests/run

.github/workflows/merge-check.yml

@@ -0,0 +1,99 @@
+name: merge-check
+
+# Why this is a second workflow rather than another job in ci.yml.
+#
+# ci.yml answers "is this pull request green?" and it already checks out the
+# MERGE RESULT — `actions/checkout` resolves `refs/pull/N/merge` on a
+# `pull_request` event, which the run for PR #16 on 2026-08-20 shows verbatim:
+#
+#     git fetch ... +9e18354...:refs/remotes/pull/16/merge
+#     HEAD is now at 9e18354 Merge 96822a4e... into e9d8c4dc...
+#
+# So "check out the merge result" is not the missing piece; it is what already
+# happens. What ci.yml cannot answer is "is the tree we are about to HAVE
+# green?", for two reasons this workflow exists to cover:
+#
+#   1. That merge result is computed against the base as it stood when the
+#      event fired, and GitHub never re-fires `pull_request` when the base
+#      branch moves underneath an open PR. PR #14 and PR #15 were both
+#      computed against e9d8c4dc; #14 landed on 02e61fd and #15 landed on top
+#      of #14, so no run ever saw the tree that shipped.
+#   2. Nothing ever tests two OPEN changes together. #14 added both store
+#      files to `claims[]`; #15 shipped tests asserting the world before that
+#      claim. Each was fine alone. Merged, two tests were red that neither
+#      change could have seen.
+#
+# Hence the trigger list below. `push` to an integration branch is exactly the
+# moment a base moves under the PRs still queued against it, and that is when
+# every one of them is re-checked.
+
+on:
+  pull_request:
+    branches: [main, feat/work-modes]
+  push:
+    branches: [main, feat/work-modes]
+  workflow_dispatch:
+
+concurrency:
+  group: merge-check-${{ github.event.pull_request.base.ref || github.ref_name }}
+  cancel-in-progress: true
+
+jobs:
+  pairs:
+    runs-on: ubuntu-latest
+    # Pair search is quadratic in the number of open PRs and only pays when the
+    # combined tree is red. This is the ceiling on that, so a bad day costs an
+    # hour and not a runner until somebody notices.
+    timeout-minutes: 60
+    permissions:
+      contents: read
+      pull-requests: read
+    steps:
+      - name: Which branch are the open PRs queued against
+        id: base
+        run: |
+          echo "ref=${{ github.event.pull_request.base.ref || github.ref_name }}" \
+            >> "$GITHUB_OUTPUT"
+
+      # The DEFAULT ref, deliberately — which on a pull_request is that PR's
+      # merge result. Two reasons. It is the only tree guaranteed to contain
+      # tests/merge-check (checking out the base cannot run a checker the base
+      # does not have yet, which is how this workflow failed on the very PR
+      # that added it). And it matches what GitHub already does with the
+      # workflow file itself on a pull_request event, so the script and the
+      # yaml driving it come from the same place.
+      #
+      # This checkout is only the source of the SCRIPT. Every tree the suite
+      # actually runs on is built by tests/merge-check in its own clone, from
+      # the base sha and the PR heads fetched below.
+      - uses: actions/checkout@v4
+        with:
+          fetch-depth: 0
+
+      # One authenticated fetch of every PR head and of the base branch.
+      # tests/merge-check clones this checkout with --shared, so it reaches
+      # these commits without credentials of its own.
+      #
+      # The destination is refs/remotes/pull/*/head, NOT refs/remotes/pull/*:
+      # actions/checkout has already written refs/remotes/pull/<n>/merge for
+      # this PR, so refs/remotes/pull/<n> is a directory and cannot also be a
+      # ref. --force does not help with a D/F collision; a sibling name does.
+      - name: Fetch every pull request head and the base branch
+        run: |
+          git fetch --no-tags --force origin \
+            '+refs/pull/*/head:refs/remotes/pull/*/head' \
+            "+refs/heads/${{ steps.base.outputs.ref }}:refs/remotes/origin/${{ steps.base.outputs.ref }}"
+
+      - uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      # Stdlib only, like the rest of Perry. `gh` is preinstalled on the runner
+      # and only used to list open PRs.
+      - name: Every open PR merged together, against the current base tip
+        env:
+          GH_TOKEN: ${{ github.token }}
+        run: |
+          python3 tests/merge-check \
+            --base "${{ steps.base.outputs.ref }}" \
+            --all-open -j 4

.gitignore

@@ -1,10 +1,23 @@
 # Auto-update timestamp written by bin/perry-update-check
 .update-check
 
+
 # macOS
 .DS_Store
 .gstack/
 
 # Python bytecode from bin/ + viewer/
 __pycache__/
 *.pyc
+
+# Subagent worktrees — temporary, created by the Agent tool
+.claude/worktrees/
+# `perry/tasks.jsonl` was ignored while it was DERIVED — `perry-tasks build`
+# reconstructed it from BOARD.md plus the event log on every call, so a tracked
+# copy would have been a second spelling of files already in the repo.
+#
+# TASK-089 made it the WRITE TARGET: `perry-task` writes the store and renders
+# the board from it. It is truth now, and an untracked truth is a file that
+# vanishes on a fresh clone while the projection of it stays. Tracked.
+
+opencode.json

.perry/config.jsonl

@@ -0,0 +1,9 @@
+{"kind": "setting", "key": "document_language", "label": "Document language", "value": "English", "order": 0}
+{"kind": "setting", "key": "chat_language", "label": "Chat language", "value": "中文", "order": 1}
+{"kind": "setting", "key": "repo_layout", "label": "Repo layout", "value": "single", "order": 2}
+{"kind": "setting", "key": "state_root", "label": "State root", "value": "perry", "order": 3}
+{"kind": "setting", "key": "pmo_repo_path", "label": "PMO repo path", "value": "/Users/bytedance/proj/Perry", "order": 4}
+{"kind": "setting", "key": "code_repo_path", "label": "Code repo path", "value": "", "order": 5}
+{"kind": "setting", "key": "last_updated", "label": "Last updated", "value": "2026-08-16", "order": 6}
+{"kind": "track", "track": "main", "mode": "project", "spine": "phase/", "stages": "", "wip": "", "sla": "", "cycle": "", "default_rung": "V3", "order": 0}
+{"kind": "track", "track": "intake", "mode": "queue", "spine": "standing", "stages": "new→triaged→in_progress→resolved", "wip": "6", "sla": "5d", "cycle": "weekly", "default_rung": "V3", "order": 1}

.perry/config.md

@@ -1,18 +1,38 @@
 # Perry configuration
 
 - Document language: English
+- Chat language: 中文
 - Repo layout: single
 - State root: perry
 - PMO repo path: /Users/bytedance/proj/Perry
 - Code repo path: —
 - Last updated: 2026-08-16
 
+## Tracks
+
+| Track | Mode | Spine | Stages | WIP | SLA | Cycle | Default rung |
+|---|---|---|---|---|---|---|---|
+| main | project | phase/ | — | — | — | — | V3 |
+| intake | queue | standing | new→triaged→in_progress→resolved | 6 | 5d | weekly | V3 |
+
+`intake` carries the work that ARRIVES — a defect an agent found mid-run, a
+sibling a sweep turned up, a review finding. It is not decomposed from a goal,
+it shows up, and its useful questions are queue questions: what has been
+waiting longest, how deep is the backlog, what keeps recurring.
+
+`main` carries the work that is DECOMPOSED — the phase, its KRs, the rows that
+serve them.
+
+Declared 2026-08-20 as the experiment in TASK-133. `Default rung` is V3 rather
+than queue mode's V2 default: an arriving row here is a code defect, and a
+resolution note is not evidence that it is fixed.
+
 ## Why the state root is not `.`
 
 Perry's own `design/` directory is the **design lane skill**
-(`design/SKILL.md`, `design/state/design_TEMPLATE.md`), not a folder of design
+(`decide/SKILL.md`, `decide/state/design_TEMPLATE.md`), not a folder of design
 documents. Pointing the state root at the project root would make Perry claim
-its own source tree, and every lint run would report `design/SKILL.md` as a
+its own source tree, and every lint run would report `decide/SKILL.md` as a
 malformed design doc.
 
 `okr/` and `pmo/` are lane skills for the same reason. `.perry/` stays at the

.perry/conformance.md

@@ -0,0 +1,38 @@
+# Perry conformance
+
+> Written by `perry-conform declare`. Each row records that **the user**
+> declared this file to match Perry's shape at that shape version
+> (ADR-004 § 4). It is a decision, not a verdict: whether the file still
+> matches is re-checked live on every write, and a row that no longer
+> holds is reported rather than deleted.
+
+> Paths are as `schema/state-schema.json` declares them — relative to the
+> state root, except the `.perry/` ones which are relative to the project
+> root. Delete a row to withdraw a declaration.
+
+| File | Shape version | Declared | Route |
+|---|---|---|---|
+| .perry/config.md | 2 | 2026-08-20 | declare |
+| .perry/hook.md | 2 | 2026-08-20 | declare |
+| BOARD.md | 2 | 2026-08-20 | declare |
+| DECISIONS.md | 2 | 2026-08-20 | declare |
+| OKR.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-001-resumable-pipelines.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-002-namespace-collision.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-003-work-modes.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-004-deterministic-writes.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-005-state-and-contracts.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-006-roles-and-knowledge.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-007-the-entity-model.md | 2 | 2026-08-20 | declare |
+| design/DESIGN-008-track-axes.md | 2 | 2026-08-28 | declare |
+| design/DESIGN-009-the-objective-is-a-record.md | 2 | 2026-08-28 | declare |
+| design/DESIGN-010-autopilot-writes-its-own-specs.md | 2 | 2026-08-28 | declare |
+| design/DESIGN-011-the-okr-is-elicited-not-collected.md | 2 | 2026-08-28 | declare |
+| knowledge/goals/linkage-graph-before-first-add.md | 2 | 2026-08-28 | declare |
+| knowledge/toolchain/pycache-staleness.md | 2 | 2026-08-20 | declare |
+| phase/001-linkage.md | 2 | 2026-08-20 | declare |
+| phase/001-work-modes-live.md | 2 | 2026-08-20 | declare |
+| phase/002-fields-are-typed.md | 2 | 2026-08-20 | declare |
+| phase/002-linkage.md | 2 | 2026-08-28 | declare |
+| phase/003-linkage.md | 2 | 2026-08-28 | declare |
+| phase/003-storage-code.md | 2 | 2026-08-28 | declare |