ethanplusai/codex-on-crack / plugins/codex-on-crack/skills/crack
crack
UnverifiedLead substantial Codex builds with one configured implementation worker and lead acceptance. Use when the user requests codex-on-crack, names a model to delegate work to, asks to run subagents through Claude, or requests a lead/worker split. Keep host-only work direct; honor explicit external model choices. Workers must not invoke this orchestrator.
references/measurement.md
View on GitHub# Controlled trials
Use fresh paired sessions with identical initial repository, prompt, root model,
effort, tools, permissions, and acceptance checks. Do not reuse a session that
has already solved the task. Keep prior lessons, global orchestration policies,
and unrelated roles out of the baseline. Do not weaken the baseline's normal
native capabilities. Record the actual models each condition uses.
Compare stock Codex, the thin workflow with same-model workers, and (only when
approved) the thin workflow with a different worker. Freeze conditions before
scoring; count failures, human intervention, and all retry/review usage. A single
pilot demonstrates operability, not improved cost or quality.
Report acceptance first, then root usage, all-session usage, time and human
intervention. Export the complete root and every descendant session privately:
`node scripts/report.mjs --plan <plan.json> --root-session <id>
--session-log <root.jsonl> --session-log <child.jsonl>`.
This counts cumulative native client snapshots once per session, includes only
linked descendants, and leaves missing usage unknown. Verify export coverage
from the host; the importer cannot discover omitted children. Provider labels
may be unavailable. Client usage is not service identity or actual billing.
`--router-log <path>` optionally provides a separate diagnostic window, never
run attribution. Native workers are absent from that log and unrelated requests
may be present. Do not use its numbers as savings evidence.
The report counts each task's final rung, not all escalation events. Retain a
separate timestamped intervention/escalation trace and recordings. Redact all
exports before publishing. Require equal held-out acceptance and report total
resources per accepted outcome; avoid lines-of-code productivity claims.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |