…it/exemption/symlink/concurrent

RED gate for Plan 02-05. 10 test functions covering:
- freeze/unfreeze persistence to freeze.json (SPRINT-02, QUALITY-06)
- is_frozen exact + path-prefix match
- Pitfall HKUDS#5 exemption: paths under get_data_dir() never vetoed
- T-02-02 mitigation: symlinks + .. segments canonicalized
- FrozenPathError message shape (D-12)
- freeze_audit.jsonl append-only (SAFETY-04, mirrors exit_journal.py)
- Pause/resume rehydration via get_freeze_registry singleton
- 8-thread concurrent writer (file_locked serialization)

All 10 tests fail with ModuleNotFoundError — Task 2 ships the implementation.

…pend-only audit

GREEN gate for Plan 02-05. Ships the sprint-scoped path write-lock registry
that Plan 02-10's BeforeFileWrite / BeforeToolCall subscribers will consult
to veto agent writes.

Implementation:
- FreezeRegistry singleton mirrors PhaseRegistry shape (§02-CONTEXT lesson HKUDS#3).
- freeze/unfreeze persist to freeze.json via file_locked + atomic_write_text
  (QUALITY-06); concurrent writers serialize with no torn state (8-thread test).
- Path canonicalization via Path.resolve(strict=False) on both store and
  compare (T-02-02 mitigation): symlinks resolve to real paths, .. segments
  collapse, relative-path bypass defeated.
- Pitfall HKUDS#5 exemption: is_frozen returns (False, "") for any path under
  get_data_dir() so sprint pause/resume never deadlocks writing its own
  state.json / freeze.json / answers/ after a user types /freeze /.
- freeze_audit.jsonl is append-only (SAFETY-04), mirroring exit_journal.py:
  one json.dumps(entry) + newline per freeze / unfreeze / freeze_glob.
- FrozenPathError subclasses ValueError so mcp.helpers.translate_error
  auto-wraps it into an MCPToolError (D-12). Message shape:
    "<path> is /freeze-locked by <who>; unfreeze via /unfreeze <path>"

Tests: 10/10 passing (test_freeze_registry.py). Regression matrix 12/12
green. Phase 1 peer tests (sprint_state, interaction_gate, phase_registry)
all green.

Plan 02-10 subscribes a handler that raises FrozenPathError when is_frozen
matches; this plan ships only the registry state-holder.

Requirements: SAFETY-02, SAFETY-04, QUALITY-06.

SUMMARY.md documents:
- Class + method line ranges (per plan <output> requirement)
- Pitfall HKUDS#5 exemption confirmation via test_is_frozen_returns_false_for_paths_under_data_dir
- T-02-02 symlink mitigation confirmation via test_is_frozen_canonicalizes_relative_paths_and_symlinks
- JSONL append-only invariant proof (5 lines after 5 actions)
- Forward contracts for Plan 02-10 / 02-11 / 02-12 / Phase 3
- 1 deviation: Rule 3 (blocking) test-fixture split for work_root vs data_root
  so Pitfall HKUDS#5 exemption and T-02-02 symlink check don't mask each other

RED commit: e0c01e6
GREEN commit: 268e423

Requirements completed: SAFETY-02, SAFETY-04, QUALITY-06

- 8 tests for Task 1: EventBus subscribers + /careful regex blacklist
- 6 tests for Task 3: clawteam guard CLI sub-app (freeze/unfreeze/guard/unguard)
- All fail with ImportError on CAREFUL_BLACKLIST + register_safety_subscribers (RED)
- Pitfall HKUDS#5 exemption re-verified at subscriber level (test 8)
- CLI tests use typer.testing.CliRunner + --json envelope assertions

Task 1 GREEN. Extends FreezeRegistry module with opt-in EventBus subscribers
that consult the registry on BeforeFileWrite + BeforeToolCall events.

- CAREFUL_BLACKLIST regex matches 5 destructive command patterns (D-13):
  rm -rf, git reset --hard, git push --force, DROP TABLE, DELETE FROM ... WHERE
- _on_before_file_write: veto writes to frozen paths (delegates to is_frozen
  which already honors the Pitfall HKUDS#5 data-dir exemption)
- _on_before_tool_call: shallow arg-scan for frozen path values
- _on_careful: warn-only by default; veto-mode via set_careful_veto_mode
  (flipped by clawteam guard guard composite)
- register_safety_subscribers(bus): idempotent opt-in registration called by
  SprintConductor.__init__ in Plan 02-11; Phase 0 templates never call this
  so their EventBus stays unsubscribed (Pitfall HKUDS#8 BC hinge)
- reset_safety_subscribers(): test helper
- 9 tests in tests/test_safety_rails.py now GREEN (Task 1 tests 1-8 plus one
  pass-through sanity check)

- 4 tests covering sprint lifecycle across Phase 2 primitives (UX-02/03/04/05/09,
  CORE-05/07, SPRINT-01/02, SKILL-09, SAFETY-02, QUALITY-01/08/11)
- Exercises CLI start/status/show/list/pause/resume via typer.CliRunner + JSON
  envelope shape
- Verifies FreezeRegistry data-dir exemption (Pitfall HKUDS#5)
- Confirms advance_phase gate chain does not crash on empty artifacts/turn-counters
- Validates AMBIGUOUS_SPRINT prefix error envelope (D-25)

…nswer/stop)

The 28-command CLI surface is overwhelming for a solo founder. This adds
4 task-oriented top-level commands that wrap the full happy path:

  clawteam go "<goal>"    — create team + start sprint + launch agents
  clawteam status          — dashboard: team/sprint/questions/cost, no tmux
  clawteam answer          — interactive question picker + $EDITOR launch
  clawteam stop            — kill tmux + cleanup team data

All four read/write an "active team" pointer at ~/.clawteam/active_team
so the user rarely passes --team explicitly. The full 28-command surface
(team, sprint, launch, attend, inbox, ...) is unchanged for power users.

Rationale:
- v1.0 UAT surfaced that the documented flow is `team spawn` + `sprint
  start` + `launch` (3 commands, easy to forget order) + `attend` +
  `attend pick <qid>` (requires knowing qid) + `team cleanup X --force`
  + `tmux kill-session -t clawteam-X` (two ways to stop).
- A solo founder shouldn't need to learn the full harness vocabulary.
- User explicitly asked for "不要做的太command heavy / 交互流程的简单
  易用 / 状态观察的直觉性" (don't make it command-heavy, prioritize easy
  interaction + intuitive state observation).

Design:
- `go`: 24-second one-shot on gstack template. Auto-names team
  `solo-<hash>` unless --name given. Shows 3-step progress (create /
  sprint / launch) + final rich Panel with next-step hints.
- `status`: single-screen view — agent count, sprint ID (truncated),
  goal, phase progress bar [█░░░░░░] think (1/7), top-5 pending
  questions (title not qid), cost panel with honest "$0.00 — 999.001
  unwired" caveat, tmux session liveness indicator (● green / ○ red).
- `answer`: numbered list of pending questions, prompt for choice,
  open question.md in $EDITOR to read, then offer to open answer.md
  for writing. Closes the loop for the HUMAN-UAT HKUDS#5 scenario.
- `stop`: interactive confirm (--force to skip), kill tmux, run team
  cleanup, clear active pointer. --keep-data variant preserves team
  data for post-mortem.

Tests: 9 unit tests covering active-team pointer roundtrip, empty-state
messaging for all 3 read commands, stale-pointer detection, stop
clearing the pointer, --team flag override, dashboard field rendering.

E2E verified on CachyOS/fish:
  clawteam go "improve README"    → 24.5s, 11 tmux windows, rich Panel
  clawteam status                  → dashboard renders correctly
  clawteam stop --force            → tmux killed, data removed, pointer cleared

Addresses v1.0 UAT finding: the primary user flow requires 10+ command
invocations spanning team/sprint/launch/attend/tmux. After this change,
90% of the flow is 4 commands.

Note: no version bump — folded into v1.0 UAT validation scope per user
decision 2026-04-22.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

User request: "用tmux hook 把prompt注入进每一轮对话 包括身份和需要用的
skill 或者用tmux hook去做一些别的事情 总之tmux hook会是一个很好的技术".

Concept: agents drift over long sessions. CEO forgets to delegate (user
just hit this bug), engineer forgets which `clawteam` subcommands are
available, shipper forgets that `/ship` lives in the framework. A
periodic re-injection of "you are X; here's what you should use"
keeps them anchored.

Implementation:

1. `clawteam nudge <pane_id>` CLI command (new clawteam/nudge.py):
   - Looks up pane_id → session → team → agent via the persisted
     tmux_pane_map.json
   - Debounces (60s cooldown per pane) so repeated silence-hook firings
     don't spam the claude TUI
   - Reads the latest sprint state (current_phase + goal) for context
   - Builds a role-specific reminder text from _ROLE_NUDGES table
     (each role gets 2-3 lines: what you ARE + what commands you CAN
     run now). Appends "Goal: X · Phase: Y" when available.
   - Injects via `tmux send-keys -l <text>` + `send-keys Enter` so
     claude processes it as a user turn.
   - Silent-fail on any error (a noisy hook drowns the user's tmux).

2. tmux hook wiring (tmux_backend.py::_configure_nudge_hook):
   - `monitor-silence = 45s` on each window
   - `alert-silence` hook fires `clawteam nudge #{pane_id}` in the
     background (`run-shell -b`)
   - Opt-out via `CLAWTEAM_NUDGE_SECONDS=0` env var
   - Called from both `TmuxBackend.tile_panes()` (happy path) and
     `TmuxBackend.enable_mouse()` (the --windows non-tiled path) so
     every team gets nudge wiring regardless of layout choice.

3. pane_map normalization (tmux_backend.py::tile_panes):
   - Post-layout readback of pane_title returned claude's dynamic
     status strings like "⠐ CEO" / "✢ ENG-MGR" (spinner glyphs prefix).
   - Normalize by matching uppercase title content against the
     pre-merge window roster (roster_upper dict). Any title containing
     a known role (e.g. "⠐ CEO" contains "CEO") maps to the canonical
     lowercase role name.
   - pane_map.json now reliably stores {"0": "ceo", "1": "pm", ...}
     regardless of what claude's currently showing.

Real-env verified:
  clawteam go --no-attach --no-window "test"
  cat ~/.clawteam/teams/<team>/tmux_pane_map.json  → clean lowercase roster
  clawteam status                                    → Panes: 0=ceo 1=pm ...
  # wait for claude to boot
  clawteam nudge --force <pane_id>                   → injects role reminder
  tmux capture-pane                                  → sees "[nudge] Reminder:
                                                      you DELEGATE, never
                                                      implement. Decompose..."

Future uses of this hook infrastructure (brainstormed but not impl):
  - HKUDS#3 pane-died → log death reason, optional respawn
  - HKUDS#5 alert-activity → append pane output to activity.jsonl timeline
  - HKUDS#6 session-created → broadcast phase+goal to all panes
  - HKUDS#7 send-keys filter → intercept dangerous commands (rm -rf, force push)
  - HKUDS#10 pipe-pane → validate agent envelope JSON against schema

Tests: 30 solo/spawn_cli still green (nudge unit tests not added yet
— next commit candidate).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>