[BUG] Loop persistence misidentified as active session despite "completed" status in progress.json

[slothvips]

.ralphrc

# .ralphrc - Ralph project configuration
# Generated by: ralph-setup
# Documentation: https://github.com/frankbria/ralph-claude-code

# Project identification
PROJECT_NAME="."
PROJECT_TYPE="generic"

# Loop settings
MAX_CALLS_PER_HOUR=100
CLAUDE_TIMEOUT_MINUTES=15
CLAUDE_OUTPUT_FORMAT="json"
RALPH_ARGS="--live --verbose"

# Tool permissions
# Comma-separated list of allowed tools
# Safe git subcommands only - broad Bash(git *) allows destructive commands like git clean/git rm (Issue #149)
ALLOWED_TOOLS="Task,Write,Read,Edit,Glob,Grep,Bash(*)"

# Session management
SESSION_CONTINUITY=true
SESSION_EXPIRY_HOURS=24

# Task sources (for ralph enable --sync)
# Options: local, beads, github (comma-separated for multiple)
TASK_SOURCES="local"
GITHUB_TASK_LABEL="ralph-task"
BEADS_FILTER="status:open"

# Circuit breaker thresholds
CB_NO_PROGRESS_THRESHOLD=3
CB_SAME_ERROR_THRESHOLD=5
CB_OUTPUT_DECLINE_THRESHOLD=70

fix_plan.md

## Medium Priority
- [x] Create UserOpControl.tsx component (start/stop/recording indicator)
- [x] Create UserOperationsList.tsx component (list, replay, edit, delete)
- [x] Create UserOperationsView.tsx (main management view)
- [x] Create useUserOperations.ts hook
- [x] Integrate components into popup with new "操作" tab
- [ ] Test recording flow in browser (manual testing required)
- [ ] Test replay flow in browser (manual testing required)

progress.json

{"status": "completed", "timestamp": "2026-02-20 17:43:47"}

Even though Ralph noted that manual testing is required, it still hasn't exited the loop. Is this a feature or a bug?

[coderabbitai]

🔗 Related PRs

#66 - feat(session): add session lifecycle management with auto-reset triggers [merged]
#90 - fix(loop): respect Claude's EXIT_SIGNAL when checking completion indicators [merged]
#137 - fix(setup): create .ralphrc with consistent tool permissions [merged]
#140 - fix: Recognize STATUS: COMPLETE as progress in circuit breaker [merged]
#158 - fix: progress detection improvements (#141, #144) [merged]

---

📝 Issue Planner

_{Check the box below or use the @coderabbitai plan command to generate an implementation plan and prompts that you can use with your favorite coding assistant.}

• Create Plan

---

🧪 Issue enrichment is currently in open beta.

You can configure auto-planning by selecting labels in the issue_enrichment configuration.

To disable automatic issue enrichment, add the following to your .coderabbit.yaml:

issue_enrichment:
  auto_enrich:
    enabled: false

💬 Have feedback or questions? Drop into our discord!

[github-actions]

Triage: This has been labeled as a bug 🐛 and assigned to @frankbria.

Analysis: The loop is incorrectly identifying a completed session as still active, likely due to a race condition or incorrect state check when reading progress.json. The session ID is being persisted and resumed even though the loop status shows STATUS: COMPLETE.

Recommended next steps:

• Investigate response_analyzer.sh — specifically the should_resume_session() function and how it validates session state against progress.json.
• Check if the RALPH_DIR/.claude_session_id file is being cleared on loop completion.
• Add a guard in session resumption logic to cross-check the loop completion state before resuming.

Generated by Issue Triage Assistant

[traycerai]

Plan

Observations

progress.json is a per-execution status file, not a project-level completion signal. It is written to "executing" at the start of each Claude invocation and "completed" after each successful one (lines 1367–1399 of file:ralph_loop.sh). The should_exit_gracefully() function (lines 493–588) never reads progress.json — it only exits when all fix_plan.md checkboxes are ticked. The two remaining items (- [ ] Test recording flow… and - [ ] Test replay flow…) are manual-only tasks Claude can never check off, so the loop runs indefinitely.

Approach

The fix is a single, targeted addition to should_exit_gracefully(): a new exit condition that detects when every remaining unchecked fix_plan.md item is annotated as requiring manual intervention. This is the minimal change that resolves the loop-persistence bug without altering any other exit path.

---

Implementation Steps

1. Add a "manual-tasks-only" exit condition to should_exit_gracefully()

In file:ralph_loop.sh, inside should_exit_gracefully(), after the existing condition #5 (the fix_plan.md all-complete check, lines 570–585), add a new condition #6:

• Read $RALPH_DIR/fix_plan.md (same guard as condition Bug: date: illegal option -- d on macOS #5).
• Count unchecked items (- [ ]) — call this uncompleted_items.
• Count unchecked items whose line text also matches a "manual" keyword pattern (e.g., case-insensitive match for manual, browser, human, manual testing) — call this manual_only_items.
• If uncompleted_items > 0 AND uncompleted_items == manual_only_items (i.e., every remaining task is manual-only), emit manual_tasks_only and return 0.

The grep pattern for manual items should be something like:

grep -cEi "^[[:space:]]*- \[ \].*\b(manual|browser|human)\b"

2. Handle the new exit reason in the main loop

In file:ralph_loop.sh, inside main(), the block at lines 1647–1698 already handles all non-empty exit_reason values with a generic graceful-exit path. The new manual_tasks_only reason will flow through this path automatically — no additional branching is needed.

Optionally, add a distinct log message for this reason to make it clear to the user why the loop stopped:

"🛑 All remaining tasks require manual intervention — exiting loop"

This message should be emitted before the existing update_status call when exit_reason == "manual_tasks_only".

3. Clarify progress.json semantics in the status output

In ralph_monitor.sh (around line 81), the monitor already reads progress.json. Add a display note when progress_status == "completed" that reads something like "Last execution: completed (loop idle)" to distinguish it from project-level completion. This prevents future confusion about what "completed" in progress.json means.

---

Flow After the Fix

flowchart TD
    A[should_exit_gracefully called] --> B{fix_plan.md exists?}
    B -- No --> Z[return empty - continue loop]
    B -- Yes --> C{All items checked?}
    C -- Yes --> D[emit plan_complete - exit]
    C -- No --> E{All remaining unchecked items\ncontain manual keyword?}
    E -- Yes --> F[emit manual_tasks_only - exit]
    E -- No --> Z

Loading

---

Summary of Changes

File	Location	Change
file:ralph_loop.sh	should_exit_gracefully() after line 585	Add condition #6: exit when all unchecked fix_plan.md items are manual-only
file:ralph_loop.sh	main() around line 1688	Add log message for manual_tasks_only exit reason
file:ralph_monitor.sh	~line 85	Clarify "completed" display label to avoid confusion with project-level completion

Import In IDE

🤖 Prompt for AI Agents

## Observations

`progress.json` is a **per-execution status file**, not a project-level completion signal. It is written to `"executing"` at the start of each Claude invocation and `"completed"` after each successful one (lines 1367–1399 of `file:ralph_loop.sh`). The `should_exit_gracefully()` function (lines 493–588) **never reads `progress.json`** — it only exits when all `fix_plan.md` checkboxes are ticked. The two remaining items (`- [ ] Test recording flow…` and `- [ ] Test replay flow…`) are manual-only tasks Claude can never check off, so the loop runs indefinitely.

## Approach

The fix is a single, targeted addition to `should_exit_gracefully()`: a new exit condition that detects when every remaining unchecked `fix_plan.md` item is annotated as requiring manual intervention. This is the minimal change that resolves the loop-persistence bug without altering any other exit path.

---

## Implementation Steps

### 1. Add a "manual-tasks-only" exit condition to `should_exit_gracefully()`

In `file:ralph_loop.sh`, inside `should_exit_gracefully()`, **after the existing condition #5** (the `fix_plan.md` all-complete check, lines 570–585), add a new condition #6:

- Read `$RALPH_DIR/fix_plan.md` (same guard as condition #5).
- Count unchecked items (`- [ ]`) — call this `uncompleted_items`.
- Count unchecked items whose line text also matches a "manual" keyword pattern (e.g., case-insensitive match for `manual`, `browser`, `human`, `manual testing`) — call this `manual_only_items`.
- If `uncompleted_items > 0` AND `uncompleted_items == manual_only_items` (i.e., every remaining task is manual-only), emit `manual_tasks_only` and `return 0`.

The grep pattern for manual items should be something like:
```
grep -cEi "^[[:space:]]*- \[ \].*\b(manual|browser|human)\b"
```

### 2. Handle the new exit reason in the main loop

In `file:ralph_loop.sh`, inside `main()`, the block at lines 1647–1698 already handles all non-empty `exit_reason` values with a generic graceful-exit path. The new `manual_tasks_only` reason will flow through this path automatically — no additional branching is needed.

Optionally, add a distinct log message for this reason to make it clear to the user why the loop stopped:

```
"🛑 All remaining tasks require manual intervention — exiting loop"
```

This message should be emitted before the existing `update_status` call when `exit_reason == "manual_tasks_only"`.

### 3. Clarify `progress.json` semantics in the status output

In `ralph_monitor.sh` (around line 81), the monitor already reads `progress.json`. Add a display note when `progress_status == "completed"` that reads something like `"Last execution: completed (loop idle)"` to distinguish it from project-level completion. This prevents future confusion about what `"completed"` in `progress.json` means.

---

## Flow After the Fix

```mermaid
flowchart TD
    A[should_exit_gracefully called] --> B{fix_plan.md exists?}
    B -- No --> Z[return empty - continue loop]
    B -- Yes --> C{All items checked?}
    C -- Yes --> D[emit plan_complete - exit]
    C -- No --> E{All remaining unchecked items\ncontain manual keyword?}
    E -- Yes --> F[emit manual_tasks_only - exit]
    E -- No --> Z
```

---

## Summary of Changes

| File | Location | Change |
|---|---|---|
| `file:ralph_loop.sh` | `should_exit_gracefully()` after line 585 | Add condition #6: exit when all unchecked `fix_plan.md` items are manual-only |
| `file:ralph_loop.sh` | `main()` around line 1688 | Add log message for `manual_tasks_only` exit reason |
| `file:ralph_monitor.sh` | ~line 85 | Clarify `"completed"` display label to avoid confusion with project-level completion |

---

Execution Information

Branch: main
Commit: 6b1b6d8

---

💡 Tips

Supported Commands (Inside Comments)

• Use @traycerai generate to iterate on the previous version of the implementation plan.

Supported Commands (Inside Description)

• Add @traycerai ignore anywhere in the ticket description to prevent this ticket from being processed.
• Add @traycerai branch:<branch-name> anywhere in the ticket description to specify the target branch for the implementation plan.

Community

• Join our Discord Community to get help, request features, and share feedback.
• Follow us on X/Twitter for updates and announcements.