| name | ralph-run-reconcile-claude-code |
| description | Audit and reconcile the latest Ralph run in Claude Code. Use when Ralph finishes with failures, tentative outcomes, or suspected missing merges; inspect story branches, state/spec drift, and merge conflicts, then produce or execute a prioritized remediation plan. |
Ralph Run Reconcile (Claude Code)
Overview
Run a deterministic post-run reconcile pass after ralph run. Detect missing merges, state drift, and run failures quickly, then apply only the fixes the user approves.
Load references/reconcile-checks.md before starting analysis.
Operating Modes
Use one mode per run:
report (default): collect evidence and produce a remediation plan without code changes.
execute-safe: apply low-risk fixes only, then verify.
execute-approved: apply low/medium/high fixes explicitly approved by the user.
Workflow
1) Capture Baseline
Run and record:
git rev-parse --abbrev-ref HEAD
git rev-parse HEAD
git status --short
date -u +"%Y-%m-%dT%H:%M:%SZ"
If the workspace is dirty, continue, but avoid reverting unrelated changes.
2) Collect Reconcile Evidence
Run deterministic checks:
ls -la .ralph 2>/dev/null || true
test -f .ralph/stories.txt && sed -n '1,220p' .ralph/stories.txt || true
test -f .ralph/state.json && jq '.completed_stories, .merged_stories, .retry_counts' .ralph/state.json || true
git for-each-ref --format='%(refname:short)' refs/heads/ralph/story-* || true
ralph status || true
ralph reconcile || true
test -f progress.txt && tail -n 120 progress.txt || true
When present, inspect run logs under ralph-run-terminal-logs/ and .ralph/logs/.
3) Build Findings Matrix
Use the checklist in references/reconcile-checks.md and produce a table with:
Finding ID
Severity
Evidence
Impact
Recommended Action
Auto-executable (yes or no)
Focus first on orphaned ralph/story-* branches with unmerged commits.
4) Classify and Plan
Map findings to risk tiers:
low: deterministic and reversible (ralph reconcile --apply, queue metadata fix, status alignment).
medium: targeted edits or retries touching implementation files.
high: merge conflicts, broad refactors, or uncertain root cause.
Always provide an ordered execution plan and explicit stop conditions.
5) Execute (Only in Execution Modes)
Apply this order:
- Run
ralph reconcile --apply if low-risk orphaned branches exist.
- Re-check state/spec/queue consistency.
- Run validation commands from
.ralph/config.json when configured.
- Re-run only the minimal set of failed stories if needed.
Never run destructive recovery commands (git reset --hard, force branch deletes, mass file deletions) unless the user explicitly asks.
6) Verify and Report
End with a concise report:
Resolved: list of findings fixed now.
Pending: remaining findings and why they were deferred.
Verification: tests/checks run and outcomes.
Next Command: exact next command for the user.
Output Contract
When done, provide:
One-line status (green/yellow/red style summary in plain text).
Findings table sorted by severity.
Execution log summary (commands actually run).
Next-step queue with concrete IDs and owners.
Cross-Model Safeguard
When the user requests extra assurance, produce a handoff prompt for Codex that includes:
- current branch + SHA
- findings table
- unresolved items
- explicit ask: "challenge these findings and list disagreements"