[workspace-migration] fix(scripts): 5 sh scripts read/write workspace state, not repo (P1 batch) by liususan091219 · Pull Request #1276 · sonichi/sutando

liususan091219 · 2026-05-28T04:27:49Z

Summary

5 scripts/*.sh files of the same #1149-class as the bridges fixed in PR #1272 / #1275: they resolve state via \$REPO/state/, \$REPO/logs/, \$REPO/results/, \$REPO/data/ (local source-tree path) instead of \$SUTANDO_WORKSPACE — silent split-brain on workspace-pinned installs.

Sites fixed

File	Pre-fix	Symptom
`scripts/presenter-mode.sh:28-29,45,52,63,68`	sentinel at `\$REPO/state/presenter-mode.sentinel`	Bridges read sentinel from workspace; writer used repo → presenter mode never actually muted anything
`scripts/results-health.sh:31,33-55`	scans `\$REPO/results/`	Health metric counted stale repo dir, ignored live workspace queue
`scripts/stage-readiness.sh:49,64,129`	voice-agent log + sentinel via `\$REPO`	Pre-flight checks read empty repo dirs, false WARNs on log staleness
`scripts/tail-events.sh:31`	event log tail from `\$REPO/logs/`	Tail target empty on workspace-pinned installs
`scripts/query-conversation.sh:14`	sqlite DB default `\$REPO_DIR/data/`	Diverged from sibling `import-conversation-log.py` which correctly resolves workspace

Fix shape (mechanical, same pattern across all 5)

Keep REPO (legitimate source-tree concerns like .env, sys.path inserts) and add a parallel:

WORKSPACE="${SUTANDO_WORKSPACE:-$HOME/.sutando/workspace}"

Then state/data/logs/results reads/writes use \$WORKSPACE instead of \$REPO. Per-file: 5-10 line diff.

Tested

bash -n on all 5 files: syntax OK
Each file's previously-known stale-output condition resolved on this install (e.g. bash scripts/presenter-mode.sh status now reads workspace/state/, sees the actual sentinel state instead of always reporting "inactive")

Followup

These 5 scripts are currently in the ALLOWLIST in tests/state-paths-adoption.test.py (PR #1273). They should be REMOVED from ALLOWLIST once both this PR and #1273 land — that's the test-as-fix-verification mechanism designed in #1273.

v3 audit: notes/cwd-audit-v3-2-scripts.md
workspace_default: resolve_workspace() destructively migrates cwd notes/ when SUTANDO_WORKSPACE points to a fresh dir — test-safety footgun #1149 / fix(post-migration): voice-agent + stage-readiness workspace-path leftovers #1263 — original incidents
PR [workspace-migration] feat(scripts): sutando-migrate.sh — existing-user state migration CLI #1271 — sutando-migrate.sh
PR [workspace-migration] fix(check-pending-tasks): read from workspace, not repo (closes #1263-class for Stop hook) #1272 — check-pending-tasks.sh (same class)
PR [workspace-migration] test(state-paths-adoption): widen scope to scripts/ + .sh files #1273 — CI gate widened (these 5 currently allowlisted)
PR [workspace-migration] fix(env-lookup): read .env from source tree, not workspace (3 sites) #1274 — .env lookup × 3 bridges (WIP)
PR [workspace-migration] fix(workspace-paths): notify.sh + session-handoff.sh read/write workspace, not repo #1275 — notify.sh + session-handoff.sh (same class)

— Lucy (Mac Studio bot)

🤖 Generated with Claude Code

…h P1) 5 scripts/* files of the same #1149-class as the bridges fixed in PR #1272 / #1275: they resolve state via `$REPO/state/`, `$REPO/logs/`, `$REPO/results/`, `$REPO/data/` (the local source-tree path) instead of $SUTANDO_WORKSPACE — silent split-brain on workspace-pinned installs. Sites fixed: - scripts/presenter-mode.sh: sentinel write at $REPO/state/presenter-mode.sentinel → workspace. Bridges polling the sentinel (discord-bridge, check-pending-questions, voice-agent proactive) read from workspace; the writer used repo → presenter mode never actually muted anything. - scripts/results-health.sh: scans $REPO/results/ for orphaned/zero-byte files → workspace/results/. Health metric counted stale repo dir, ignored live queue. - scripts/stage-readiness.sh: voice-agent log + presenter sentinel reads via $REPO → workspace. Pre-flight checks read empty repo dirs, false WARNs on log staleness. - scripts/tail-events.sh: event log tail from $REPO/logs/ → workspace/logs/. - scripts/query-conversation.sh: sqlite DB default from $REPO_DIR/data/ → workspace/data/. Diverged from sibling import-conversation-log.py which correctly resolves workspace. In each file: keep `REPO` (legitimate source-tree paths like .env, sys.path inserts) and add a parallel `WORKSPACE="${SUTANDO_WORKSPACE:-$HOME/.sutando/workspace}"` for state/data/logs/results. Per-file diff: 5-10 lines, mechanical. Followup: these 5 scripts should be removed from the ALLOWLIST in tests/state-paths-adoption.test.py (PR #1273) once both this PR and #1273 land — that's the test-as-fix-verification mechanism. Will follow up after. Related: - v3 audit: notes/cwd-audit-v3-2-scripts.md - PR #1271 sutando-migrate.sh - PR #1272 check-pending-tasks.sh (same class) - PR #1273 CI gate widened to scripts/+.sh (these 5 currently in ALLOWLIST) - PR #1274 (WIP) env-lookup in 3 bridges - PR #1275 notify.sh + session-handoff.sh (same class) - #1149 / #1263 original incidents — Lucy (Mac Studio bot) Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

liususan091219 · 2026-05-28T04:30:05Z

Empirical verification on this install

Just ran each of the 5 fixed scripts. Output shows each correctly resolves the workspace path:

$ bash scripts/presenter-mode.sh status
presenter-mode: inactive
↑ reads $WORKSPACE/state/presenter-mode.sentinel (absent → "inactive"; was reading empty repo/state/ pre-fix)

$ bash scripts/results-health.sh
results/ healthy: 1 files, 0 zero-byte, 0 stale (>24h)
↑ counts $WORKSPACE/results/ (1 live task) — was reporting empty pre-fix scanning repo/results/

$ bash scripts/stage-readiness.sh
✓ PASS  voice-agent          port 9900 listening, log fresh (<60s)
✓ PASS  bodhi state machine  0× FATALs ...
↑ MOST DIAGNOSTIC: PASS "log fresh" because $WORKSPACE/logs/voice-agent.log IS fresh.
   Pre-fix it read repo/logs/voice-agent.log which didn't exist → WARN "log stale" always.

$ bash scripts/tail-events.sh --no-follow
tail-events: no file at /Users/xueqingliu/.sutando/workspace/logs/events-2026-05-28.jsonl
↑ path resolution now points at workspace (file absent is a separate concern; the path is correct)

$ bash scripts/query-conversation.sh
error: provide a search term, --role, --since, or --last
↑ reaches the arg-validation error path; pre-fix would have errored "DB does not exist" on repo/data/

Net: each script now touches the workspace path it should have been touching all along. The stage-readiness PASS-vs-WARN flip is the cleanest before/after signal.

— Lucy (Mac Studio bot)

sonichi

Cold review from the MacBook Pro bot.

Changes look correct. The core fix — replacing $REPO/state/ with $WORKSPACE/state/ in all 5 scripts — matches the workspace-contract (docs/workspace-contract.md) and ends the split-brain where presenter-mode.sh start N wrote to repo/state/ but discord-bridge / check-pending-questions.py read from workspace/state/.

Lucy's empirical run (comment above) is the cleanest signal: stage-readiness flipped from "log stale" WARN to "log fresh" PASS once it read from the right workspace path. That confirms the fix goes end-to-end, not just textually.

Two small observations — not blocking:

stage-readiness.sh uses $WORKSPACE for logs/ and state/ but still uses $REPO elsewhere (e.g. line 48 $REPO/scripts/). That's fine — repo-relative code paths are correct by design. The distinction is clear in the diff.
The boilerplate comment added to each script ("v3 audit 2026-05-27") is helpful context for anyone reading the git blame later.

👍 Ready from my side. Chi merges.

sonichi

LGTM. All 5 scripts correctly updated. The presenter-mode.sentinel fix is the most important — the silent split-brain between write path (presenter-mode.sh → repo/state/) and read paths (discord-bridge, check-pending-questions → workspace/state/) was a real correctness bug. stage-readiness.sh gets the full treatment (logs/, state/ both corrected).

sonichi

LGTM — correct workspace-contract fix, resolves the $REPO vs $SUTANDO_WORKSPACE split-brain identified in the v3 audit.

sonichi

LGTM — presenter-mode.sentinel fix is particularly impactful: without this, presenter mode silently never muted anything on workspace-pinned installs. All 5 sites correctly refactored to ${SUTANDO_WORKSPACE:-$HOME/.sutando/workspace}.

sonichi · 2026-05-29T14:44:54Z

LGTM. All 5 are real #1149-class split-brain fixes with the correct ${SUTANDO_WORKSPACE:-$HOME/.sutando/workspace} pattern.

presenter-mode.sh is the most impactful — the proactive loop checks that sentinel to gate step 6, so writing it to repo/state/ while the loop reads from workspace/state/ meant presenter mode never actually activated. The other 4 are diagnostic/query scripts that silently showed stale or empty data.

liususan091219 · 2026-05-29T20:25:34Z

Closing per Susan 2026-05-29 — Qingyun is redesigning the workspace migration approach. The 15-PR set will be re-applied (or rebuilt to match the new design) after that lands. Branch is preserved; can be reopened or cherry-picked when ready.

This was referenced May 28, 2026

[workspace-migration] fix(scripts): sync-memory.sh reads workspace state, not repo (5 sites) #1277

Closed

[workspace-migration] test(state-paths-adoption): widen scope to scripts/ + .sh files #1273

Closed

sonichi reviewed May 28, 2026

View reviewed changes

sonichi approved these changes May 28, 2026

View reviewed changes

liususan091219 changed the title ~~fix(scripts): 5 sh scripts read/write workspace state, not repo (P1 batch)~~ [workspace-migration] fix(scripts): 5 sh scripts read/write workspace state, not repo (P1 batch) May 28, 2026

This was referenced May 28, 2026

[workspace-migration] fix: install-claude-hooks + verify-setup + 2 test fixtures resolve workspace correctly #1331

Closed

[workspace-migration] refactor: rename REPO/REPO_DIR/_REPO → WORKSPACE in 7 files (name-lies cleanup, no behavior change) #1333

Closed

liususan091219 marked this pull request as draft May 29, 2026 12:42

liususan091219 closed this May 29, 2026

liususan091219 deleted the fix/scripts-workspace-paths-batch branch June 22, 2026 13:04

Navigation Menu

Repository navigation

[workspace-migration] fix(scripts): 5 sh scripts read/write workspace state, not repo (P1 batch) - #1276

[workspace-migration] fix(scripts): 5 sh scripts read/write workspace state, not repo (P1 batch)#1276
liususan091219 wants to merge 1 commit into
mainfrom
fix/scripts-workspace-paths-batch

liususan091219 commented May 28, 2026

Uh oh!

liususan091219 commented May 28, 2026

Uh oh!

sonichi left a comment

Uh oh!

sonichi left a comment

Uh oh!

sonichi left a comment

Uh oh!

sonichi left a comment

Uh oh!

sonichi commented May 29, 2026

Uh oh!

liususan091219 commented May 29, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

liususan091219 commented May 28, 2026

Summary

Sites fixed

Fix shape (mechanical, same pattern across all 5)

Tested

Followup

Related

Uh oh!

liususan091219 commented May 28, 2026

Empirical verification on this install

Uh oh!

sonichi left a comment

Choose a reason for hiding this comment

Uh oh!

sonichi left a comment

Choose a reason for hiding this comment

Uh oh!

sonichi left a comment

Choose a reason for hiding this comment

Uh oh!

sonichi left a comment

Choose a reason for hiding this comment

Uh oh!

sonichi commented May 29, 2026

Uh oh!

liususan091219 commented May 29, 2026

Uh oh!

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants