juanandresgs/claude-ctrl

Codex Result Handling guides how to present Codex helper output back to the user without changing its meaning. Use it for review results, failures, setup issues, and any output that must preserve findings, severity, and boundaries.

Codex Runtime forwards a single rescue request to the codex-companion runtime from Claude Code. Use it in the codex:codex-rescue subagent when you need diagnosis, research, planning, or a fix request sent through one task call.

Context Preservation generates structured session summaries before compaction so work can resume without losing goals, changes, decisions, or next steps. Use it when a Claude Code session is about to be compacted or needs a handoff.

Decide generates interactive configurators for complex trade-off decisions from research, plans, or conversation context. It is used when options have costs, dependencies, and multiple criteria to compare.

Deep Research queries OpenAI, Perplexity, and Gemini in parallel for a topic and compares their findings. Use it when you need a synthesized research summary with agreements, disagreements, and unique insights.

GPT-5.4 Prompting gives rules and reusable blocks for writing compact Codex and GPT-5.4 prompts for coding, review, diagnosis, and research tasks. Use it when you need to structure a prompt with clear output, verification, and grounding constraints.

Observatory analyzes hook, agent, and dispatch health, then summarizes patterns, trends, and suggestion status. Use it to review operational metrics and manage the improvement loop.

PRD writes structured product requirements documents and feature specs with problem statements, goals, journeys, requirements, and success metrics. Use it when defining what to build, why it matters, and how to measure success.

Reckoning analyzes a project's MASTER_PLAN.md to assess coherence, intent alignment, and how the plan has evolved. Use it to compare trajectory, spot scope drift, or turn findings into actionable work.

Signal Trace traces broken signals through hooks, state, and dispatch paths to find where data breaks, races, or turns to garbage. Use it when an agent reads the wrong context, output goes stale, or a systemic bug is blamed on behavior.