Skip to content

oprogramadorreal/optimus-claude / skills/gauntlet

gauntlet

Verified

Runs a Gauntlet Loop: turns an ambitious goal into a builder/critic prompt judged against a concrete quality bar, confirms with the user, then leads the run until the output beats the bar or the user stops it — or exports a paste-ready /goal prompt for a fresh session. Use for long-horizon goals with an inspectable reference. Long-running; spawns many subagents, edits project files, and commits and pushes finished pieces to a feature branch.

FileKindSize
README.mdmarkdown13.0 KB
SKILL.mdskill md11.1 KB
agents/openai.yamlconfig198 B
references/goal-handoff-claude.mdmarkdown1.7 KB
references/goal-handoff-codex.mdmarkdown3.9 KB
references/goal-handoff.mdmarkdown4.2 KB

references/goal-handoff-codex.md

View on GitHub
# Codex goal handoff

Shared export, completion, and size rules live in the calling
`goal-handoff.md` reference.

## Opening instruction

Add a compact instruction to use one native Codex goal for the entire run,
following the runtime instructions in the handoff page, and to mark it
complete only after verifying the shared completion evidence. Do not carry
Claude's small-evaluator assumptions or ultracode recommendation into it.

When native goals may be unavailable in the destination, also print a
separately labeled fallback block: the same body with `/goal ` replaced by
"Plain fallback: run in this session without a native goal; nothing
continues across turns automatically." and this opening instruction removed.

## Runtime instructions

Include these in the handoff page for the destination lead:

- Inspect the current native goal using available controls (`get_goal` when
  exposed). Reuse the goal created by the pasted message. If none exists,
  this explicit goal request authorizes creating it with `create_goal` when
  available; use the full objective and completion condition. Do not create
  a goal per piece or replace an unrelated unfinished goal.
- Confirm native goal activation before dispatching work. If the destination
  lacks the required controls, explain that no persistent goal was started
  and stop; never silently claim an ordinary run has goal continuation. A
  message that starts "Plain fallback" skips the goal steps and runs in this
  session, saying that nothing continues across turns automatically.
- Use `update_goal` only as its current contract permits. Native goal
  status is not a critic. Follow the host's blocked-state threshold; do not
  equate difficult work or a plateau with an impasse while actionable work
  remains.
- Set a token budget only when the user explicitly requested one. Honor
  host pause, usage, budget and permission controls without declaring the
  goal complete or inventing tool operations to bypass them. Preserve the
  unresolved gaps and next action in the checkpoint when suspended.
- Use fresh critic contexts (`fork_turns: "none"` when supported) and
  sequence dispatches to fit available agent slots. On a new-session resume,
  rebuild each builder's context from the checkpoint; do not assume live
  agents transfer. Reopen the bar and verify saved evidence before continuing.

## Checklist above the message

- A NEW Codex session in the run's directory, with the handoff page and all
  bar materials present. Select the requested model in the host; for Astra,
  select GPT-6 Astra. Use Codex's own reasoning controls, not Claude effort
  commands. Check the sandbox and permissions needed by the run.
- Verify that native `/goal` is available. If absent, check the host/version
  and `features.goals` setting; enable it through the host only if desired;
  otherwise paste the labeled plain fallback instead, which runs without
  cross-turn continuation.
- Paste the block, then inspect `/goal` or the goal progress row to confirm
  the full objective and final completion condition are active. Correct a
  truncated objective before letting the run continue.
- Use `/goal pause`, `/goal resume`, `/goal edit`, or `/goal clear`, or the
  corresponding progress-row controls. A budget limit or blocked/paused
  state is not success. Resume with the checkpoint available; a different
  session needs the exported instructions and files again.
- After the run, in the run's session: `$optimus:commit` for any leftovers,
  then `$optimus:pr`; then `$optimus:code-review` in a fresh conversation.

Sources: [Codex goals](https://learn.chatgpt.com/use-cases/follow-goals),
[goal controls and size limit](https://learn.chatgpt.com/docs/developer-commands?surface=cli),
and [long-running work and permissions](https://learn.chatgpt.com/docs/long-running-work).
Native tool names above are conditional on the destination's exposed tools;
their current contracts take precedence over assumptions about another host.
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: 471kUpdated Claude CodeCodexVerifiedInstall
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: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
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 availableStars: 322kUpdated Claude CodeCodexVerifiedInstall