fix: auto mode gets stuck in validation loop on pre-existing test failures

[rubenmarcus]

Summary

Three targeted fixes for the auto mode validation loop issue:

1. Skip test commands in auto mode — only run build/lint validation. Pre-existing test failures (e.g., missing jsdom dep) would cause infinite loops where the agent keeps trying to fix unrelated tests. Tests remain available via --validate in non-auto mode.

2. Clean .ralph/ between batch tasks — prevents stale iteration logs and activity data from previous tasks contaminating the next task's context.

3. Set skipPlanInstructions in auto mode — resolves conflicting instructions where context-builder says "study IMPLEMENTATION_PLAN.md" but task-executor says "do NOT implement items from IMPLEMENTATION_PLAN.md".

Root Cause

When auto mode runs npm test and tests fail due to pre-existing structural issues (missing dependencies, wrong test environments), the agent enters an infinite loop trying to fix unfixable tests instead of moving on after completing the implementation.

Test plan

• Run ralph-starter auto on a project with pre-existing test failures — should complete tasks without looping
• Verify build/lint validation still runs in auto mode
• Verify .ralph/ is clean at the start of each task
• Verify no conflicting IMPLEMENTATION_PLAN.md instructions in auto mode

Closes #209

🤖 Generated with Claude Code

[github-actions]

✔️ Bundle Size Analysis

Metric	Value
Base	1848.54 KB
PR	1850.44 KB
Diff	1.90 KB (0%)

Bundle breakdown

156K	dist/auth
28K	dist/automation
4.0K	dist/cli.d.ts
4.0K	dist/cli.d.ts.map
16K	dist/cli.js
12K	dist/cli.js.map
460K	dist/commands
28K	dist/config
4.0K	dist/index.d.ts
4.0K	dist/index.d.ts.map
4.0K	dist/index.js
4.0K	dist/index.js.map
536K	dist/integrations
84K	dist/llm
732K	dist/loop
172K	dist/mcp
32K	dist/presets
92K	dist/setup
36K	dist/skills
392K	dist/sources
76K	dist/ui
84K	dist/utils
336K	dist/wizard

[greptile-apps]

Greptile Summary

This PR addresses the auto mode validation loop issue with three targeted fixes. The changes skip test commands during auto mode validation (keeping only build/lint checks), clean the .ralph/ directory between batch tasks to prevent stale context contamination, and set skipPlanInstructions: true to resolve conflicting instructions about IMPLEMENTATION_PLAN.md.

Key Changes:

• executor.ts:492-496 - Filter out test commands from validation in auto mode to prevent infinite loops when pre-existing test failures (like missing dependencies) are unfixable
• task-executor.ts:106-110 - Clean .ralph/ directory between tasks to avoid stale iteration logs and activity data
• task-executor.ts:146 - Set skipPlanInstructions: true to eliminate conflicting instructions where context-builder says "study IMPLEMENTATION_PLAN.md" but task-executor says "do NOT implement items from IMPLEMENTATION_PLAN.md"

The implementation is clean, focused, and directly addresses the root cause without over-engineering.

Confidence Score: 5/5

• This PR is safe to merge with minimal risk
• The changes are well-scoped surgical fixes targeting a specific infinite loop bug. Each change has clear rationale: filtering test commands prevents unfixable test failures from looping, cleaning .ralph/ prevents context pollution, and skipPlanInstructions eliminates instruction conflicts. The modifications are minimal, non-breaking, and follow existing patterns in the codebase
• No files require special attention

Important Files Changed

Filename	Overview
src/loop/executor.ts	Added test command filtering in auto mode to prevent infinite loops on pre-existing test failures (lines 492-496)
src/loop/task-executor.ts	Added .ralph/ directory cleanup between batch tasks and set skipPlanInstructions: true to avoid conflicting instructions

_{Last reviewed commit: 5707243}

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 5707243592

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 77040cbfc1

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".