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/replay.md
View on GitHub# Replay: registering sources explicitly
The viewer ships inside this skill at `viewer/`. It reads only the sources you
name on the command line and serves them on `127.0.0.1` behind a per-session
token. It never scans for sessions on its own.
## One view for host, worker, and external lead
```sh
VIEW=<this skill directory>/viewer
# External lead runs (a run directory, live or finished)
node "$VIEW/bridge.mjs" --run <run-dir> --out .local/replay
node "$VIEW/serve.mjs" --run <run-dir> [--run <run-dir>]
# Codex host and native worker sessions (their own jsonl files)
node "$VIEW/serve.mjs" --session <host-session.jsonl> --session <worker-session.jsonl>
# One view with everything, so the relationship is visible at once
node "$VIEW/serve.mjs" --session <host-session.jsonl> --session <worker-session.jsonl> --run <run-dir>
# A previously exported normalized event file
node "$VIEW/serve.mjs" --events <normalized.jsonl>
```
`--root <agent-id>` selects the root when a session file is ambiguous; by
default the first `--session` file is the root, and descendants are included
through their recorded parent links.
## Where the session files come from
Codex records session jsonl per session; pass the files belonging to the host
turn and to each native worker explicitly. Do not paste a directory: the viewer
has no directory-scanning mode on purpose. If you do not know which files belong
to the turn, register the host session first, confirm it appears in the view,
then add each worker file in turn.
## What the public view shows, and what it never shows
Shown: agent role label (project lead / project lead · solo / worker), model
label, status, observed span from event timestamps, tool activity, plan step
counts, and recorded token/cache counters. The final activity line for each run
summarises the tool mix and the recorded usage, including a clearly labelled
list-price equivalent that is not a charge.
Never shown: transcripts, prompts, tool arguments or outputs, file paths, tool
ids, invocation/request/session ids, phases, or billing strings. Plan text is
replaced with generic labels by default. Add `--details` to opt in to local
plan, assignment, and commentary text from native sources. This can reveal
project information despite pattern redaction; review it before a demo. Use
the UI safe-export action for sharing, rather than sharing the live state.
## Usage semantics
- One session-scope record per model per run, from the authoritative cumulative
totals. Successive runs of the same Claude session replace the earlier total;
they are never summed.
- A different Claude session is a different agent, so two sessions can never
overwrite each other's totals.
- Per-invocation deltas stay in the local `usage-summary.json` (and in the run's
own `resume-delta.json`).
- Missing counters stay missing: the view says so instead of inventing zeros.
- Account quota is not observable here, and no allowance or savings claim is
made.
## Offline and live
Replay works the same offline (from the exported bundle) and live: a run that is
still in flight shows a provisional state, and the view refreshes as the source
files grow. `bridge.mjs --out <dir>` writes `replay.jsonl`, `replay.json`, and
`usage-summary.json`; the bundle is redacted and safe to import into the UI or
share.
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 |