skills/ralph-run-reconcile-codex/SKILL.md
Audit and reconcile the latest Ralph run in Codex-driven repositories. Use when Ralph finishes and you want a post-run diagnostic or remediation pass to catch unmerged story branches, failures, state/spec drift, and merge issues, then execute a bounded recovery plan.
npx skillsauth add becerra-alberto/ralphetamine ralph-run-reconcile-codexInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
Run a structured reconcile phase after ralph run to detect and fix post-run gaps. Prioritize deterministic checks before edits, then apply bounded fixes with clear verification.
Load references/reconcile-checks.md before starting analysis.
Use one mode per request:
report (default): findings and action plan only.execute-safe: apply low-risk actions automatically.execute-approved: apply only explicitly approved medium/high actions.Collect:
git rev-parse --abbrev-ref HEAD
git rev-parse HEAD
git status --short
date -u +"%Y-%m-%dT%H:%M:%SZ"
Treat existing user changes as immutable unless asked otherwise.
Run:
rg --files .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
If available, inspect .ralph/logs/ and ralph-run-terminal-logs/ for failure signatures.
Use the checklist in references/reconcile-checks.md.
Output columns:
IDSeverityArtifactProblemActionExec Mode (safe, approved, manual)Sort by severity, then by user impact.
Create an ordered plan with rollback notes:
Include exact commands and expected post-condition for each step.
In execute modes, apply:
ralph reconcile --apply when orphaned branches are clean-merge candidates.Respect strict boundaries:
Always finish with:
ralph reconcile result.ralph/config.jsonDeliver:
Status summary (1 line)Findings matrix (severity sorted)Actions executed (command list)Verification resultsRemaining queueWhen requested, generate a Claude Code challenge prompt that asks Claude to dispute the findings matrix and highlight disagreements before any high-risk execution.
development
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.
testing
Run the complete Ralphetamine pipeline interactively: PRD creation, multi-perspective Review Party gates, spec generation, premortem review, and run script generation. Pauses for user input at key decision points. Triggers on: /ralph-pipeline-interactive, interactive ralph pipeline, ralph guided pipeline.
devops
Run the complete Ralphetamine pipeline fully autonomously with zero user input: PRD creation, spec generation, double premortem review, and run script generation. Triggers on: /ralph-pipeline-full-auto, ralph autopilot, ralph full auto, ralph autonomous pipeline, ralph auto.
tools
Convert a PRD into Ralphetamine story specs and queue. Reads a PRD file and generates specs/epic-{N}/story-{N.M}-{slug}.md files plus .ralph/stories.txt. Triggers on: /ralph, convert prd to specs, generate ralph specs, create story specs.