luna-maxing
Skill1 starsChanged 57 days agoApache-2.0View on GitHub
Use when the user asks for Luna-maxing, or to build an epic, multi-milestone plan or stacked set of PRs with Codex agents (gpt-5.6-luna, max effort) doing the implementation and review while Claude only orchestrates; also when resuming such a build after a context reset, a Codex usage limit, or a model capacity error.
In this repositoryOne other file it gives its agents.
- autonomous-loopSkill
The file
We split this skill into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 54
Every seat runs in the background.
- Distinctive: line 56
Every decision taken on the user's behalf is a numbered ledger ruling with its cost if wrong, and is carried into the next dispatch and the …
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Luna-maxing
Overview
Tandem build: Claude orchestrates, Codex builds and reviews. Every line of product code, every fix and every review verdict comes from a Codex seat (codex exec -m gpt-5.6-luna, effort max). The Claude session plans, rules on ambiguities, dispatches, runs the local gates itself, and owns git and draft PRs. Nothing is pushed until the controller's own gates are green on that exact head AND the Codex review is clean.
REQUIRED SUB-SKILL: superpowers:subagent-driven-development (ledger, briefs, review packages, fix-loop discipline). This skill replaces its subagents with Codex seats.
Roles
| Seat | Who | Session |
| Orchestrator: plan, rulings, dispatch, gates, git, rebases without conflicts, draft PRs, ledger | Claude | this session |
| Implementer, fixer, conflict-resolving rebase job | Codex Luna max | implementer thread, resumed for fixes |
| Reviewer, scoped re-reviewer | Codex Luna max | fresh session every time |
The controller never edits product code. A controller "quick fix" skips review.
Quick reference
| Step | Tool |
| Dispatch a seat (background) | LUNA_WS=<ws> scripts/dispatch.sh NAME WORKTREE PROMPT (--resume THREAD for fixes) |
| Wait out a usage limit | scripts/wait-codex-reset.sh "<RESET_AT>" (background) |
| Local green | LUNA_WS=<ws> LUNA_GATES_PROFILE=profiles/<project>.sh scripts/gates.sh NAME WORKTREE BASE [args] |
| Prompt bodies | templates/implementer.md, reviewer.md, rereview.md |
| Prompt headers, fix round, rebase job, continuation | templates/headers.md, fix-round.md, rebase-job.md, continue.md |
| Constraints file (worked example), seed lessons | templates/constraints.md (examples/constraints-example-saas.md), templates/lessons.md |
| Gate profiles | profiles/example-fullstack.sh (full example), profiles/example-minimal.sh (template), profiles/autonomous-loop.sh (repository gate) |
| Setup, loop, stacking, failure playbook | reference.md |
Paths are relative to this skill's directory. Scripts and templates are the working versions; adapt headers per task, never the review rubric.
The loop (per task)
- Worktree per task branch; record BASE.
- Implementer dispatch = header (brief path, constraints path, spec sections, worktree, report path, controller notes, lessons) + implementer body.
- On completion: reviewer (fresh) and gates in parallel.
- Findings: rule in the ledger where needed; resume the implementer thread with the findings verbatim; scoped re-review of the fix diff plus gates. Repeat until clean.
- Push, open a DRAFT PR stacked on the previous task's branch, ledger the completion.
Details, stacking and rebase handling: reference.md.
Standing rules
- At most two Codex seats run at once, each in its own worktree.
- Every seat runs in the background; the harness notifies on exit. Never poll in the foreground.
- Prompts carry file paths, not pasted content; reports and reviews come back as files.
- Every decision taken on the user's behalf is a numbered ledger ruling with its cost if wrong, and is carried into the next dispatch and the re-review header.
- Gates include suites that only run after merge, compare pre-existing failures structurally, confirm targeted tests ran rather than skipped, and check the tree is clean at start and end.
- Draft PRs only; never merge; follow the user's attribution and PR-writing rules (check commit bodies in gates).
- Acceptance that needs a deployed environment is an open gate listed in the PR body, never simulated.
- Usage limit or capacity error: wait and re-dispatch the same model (see the playbook); do not hand the seat to another model.
Red flags
| Thought | Reality |
| "Faster if I fix this one line myself" | Unreviewed controller code. Resume the implementer. |
| "Codex is rate-limited, Claude can take this review" | The user chose Codex seats. Wait for the reset. |
| "The implementer says tests pass" | Claims are not gates. Run gates.sh on the exact head. |
| "PR CI is green enough" | Post-merge-only workflows exist. Read the triggers. |
| "Review aborted halfway, findings so far are enough" | An aborted review wrote nothing. Re-dispatch fresh. |
| "Rebase conflict is trivial, resolve it here" | Conflict resolution is code. Codex rebase job, then range-diff review. |
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.